From ca34d7a7c77fa8dc823fd05439870e5a96610c5a Mon Sep 17 00:00:00 2001 From: Ulrich Drepper Date: Thu, 14 Aug 1997 01:54:13 +0000 Subject: Update. 1997-08-14 03:14 Ulrich Drepper * elf/dl-minimal.c: Don't use relative #include paths. (malloc): Don't try to initialize _dl_pagesize, it already is. (__sigjmp_save, longjmp): Reformat. * sysdeps/generic/dl-sysdep.c: Implement _dl_show_auxv function to show content of auxiliary array. (_dl_sysdep_start): Remember start of auxiliary vector. * sysdeps/mach/hurd/dl-sysdep.c: (_dl_show_auxv): New dummy function. * elf/link.h: Add prototype for _dl_show_auxv. * elf/rtld.c (dl_main): Call _dl_show_auxv if LD_SHOW_AUXV envvar is present. * libio/iofdopen.c: Better prepare for use in libstdc++. * libio/iofflush.c: Likewise. * libio/iofgetpos.c: Likewise. * libio/iofgets.c: Likewise. * libio/iofopen.c: Likewise. * libio/iofputs.c: Likewise. * libio/iofread.c: Likewise. * libio/iofsetpos.c: Likewise. * libio/ioftell.c: Likewise. * libio/iofwrite.c: Likewise. * libio/iogetdelim.c: Likewise. * libio/iogets.c: Likewise. * libio/iopopen.c: Likewise. * libio/ioputs.c: Likewise. * libio/iosetbuffer.c: Likewise. * libio/iosetvbuf.c: Likewise. * libio/ioungetc.c: Likewise. * libio/iovsprintf.c: Likewise. * libio/iovsscanf.c: Likewise. * manual/creature.texi: Mention _XOPEN_SOURCE_EXTENDED in correct place. * manual/math.texi: Add comment to sincos decsription to say it's a GNU extension. Fix description of random function. * posix/fnmatch.h: Recognize Windog by _WINDOWS32 not WIN32. * posix/glob.c: Likewise. Handle stupid system headers on SunOS. Add casts for all __alloca calls. * posix/glob.h: Recognize Windog by _WINDOWS32 not WIN32. Declare additional GLOB_* constants also if _GNU_SOURCE is defined. Patches by Paul D. Smith . * stdio-common/_itoa.c: Don't use relateive #include paths. * sysdeps/i386/fpu/bits/mathinline.h: Enable optimizations even for gcc 2.7. (atan): Loading 1.0 from memory does not pay off. (asinh): Fix typo. 1997-08-13 09:44 Philip Blundell * sysdeps/stub/accept.c: Use socklen_t where necessary. * sysdeps/stub/bind.c: Likewise. * sysdeps/stub/connect.c: Likewise. * sysdeps/stub/getsockname.c: Likewise. * sysdeps/stub/getsockopt.c: Likewise. * sysdeps/stub/recvfrom.c: Likewise. * sysdeps/stub/sendto.c: Likewise. * sysdeps/stub/setsockopt.c: Likewise. 1997-08-11 Andreas Schwab * io/ftw.c (object_compare): Avoid memcmp when comparing scalar types. 1997-08-12 Andreas Schwab * math/libm-test.c (cbrt_test): Add epsilon for long double to `cbrt (-0.001)'. 1997-08-11 Andreas Schwab * sysdeps/stub/bits/string.h: Fix comment. 1997-08-01 Andreas Schwab * elf/Makefile (routines): Remove dl-version. (dl-routines): Add it here instead. This removes dead code from libc.so. 1997-08-12 17:17 Wolfram Gloger * malloc/thread-m.h (thread_atfork, thread_atfork_static): New macros to encapsulate pthread_atfork's functionality. * malloc/malloc.c (ptmalloc_lock_all, ptmalloc_unlock_all): New functions, registered via thread_atfork in the initialization routine or via thread_atfork_static in global scope. This ensures a consistent state of all locks across fork(). 1997-08-11 23:00:32 1997 Mark Kettenis * login/programs/xtmp.c (compare_entry): Partially undo patch made by Klaus Espenlaub. There was no xtmp/utmp typo. Added comment to cclarify this. * login/programs/xtmp.h: Likewise. * posix/getopt.c: Recognize Windog by _WINDOWS32 not WIN32. Patch by Paul D. Smith . 1997-08-12 13:51 Ulrich Drepper * login/getutid.c: Rename to __getutid and make getutid and getutxid weak aliases. Patch by ir. Mark M._Kettenis . 1997-08-11 23:55 Ulrich Drepper * sysdeps/unix/sysv/linux/Makefile [subdir=misc] (sysdep_headers): Add bits/mman.h. 1997-08-11 08:51:21 1997 H.J. Lu * stdlib/fmtmsg.c (fmtmsg): Add static to __libc_once_define macro. * libc.map (__progname_full, __progname): Added. 1997-08-04 08:31 H.J. Lu * sysdeps/unix/sysv/linux/if_index.c (if_nametoindex): Return 0 if SIOGIFINDEX is not defined. (if_indextoname): Initialize "result" to NULL. (if_nametoindex, if_nameindex): Fix ENOSYS setting. 1997-08-11 11:58 Richard Henderson Sparc ISO C 9x exception handling: * sysdeps/sparc/fpu/bits/mathdef.h: New file. * sysdeps/sparc/fpu/fclrexcpt.c: New file. * sysdeps/sparc/fpu/fegetenv.c: New file. * sysdeps/sparc/fpu/fegetround.c: New file. * sysdeps/sparc/fpu/feholdexcpt.c: New file. * sysdeps/sparc/fpu/fesetenv.c: New file. * sysdeps/sparc/fpu/fesetround.c: New file. * sysdeps/sparc/fpu/feupdateenv.c: New file. * sysdeps/sparc/fpu/fgetexcptflg.c: New file. * sysdeps/sparc/fpu/fraiseexcpt.c: New file. * sysdeps/sparc/fpu/fsetexcptflg.c: New file. * sysdeps/sparc/fpu/ftestexcept.c: New file. * sysdeps/sparc/sparc32/fpu/bits/fenv.h: New file. * sysdeps/sparc/sparc64/fpu/bits/fenv.h: New file. 1997-08-11 11:58 Richard Henderson Sparc64 merge: * configure.in: Change up subdirectories, rename sparc8->sparcv8. * csu/Makefile: Nuke need-nopic-initfini bits. The problem was that sparc64 ld did not recognize pic relocs at all. * elf/dl-deps.c: Include . * elf/do-rel.h (elf_dynamic_do_rel): Take reloc address and size directly instead of the DT names to get them from. * elf/dynamic-link.h (_ELF_DYNAMIC_DO_RELOC): New macro distilled from ELF_DYNAMIC_DO_REL{,A}. Define a second version that accounts for the possibility that DT_RELA and DT_JMPPLT overlap, as such is the case on the Sparc. * elf/elf.h: Stub out Elf64_Ver*, and make them the same as the Elf32_Ver* bits. The linker does not even know to differentiate between the word sizes. Someone should examine this and figure out what the proper sizes of things should be for 64-bit hosts before the sparc64 libraries are distributed. * shlib-versions: Rename sparc64 ld.so to not conflict with sparc32 on the same machine. Move old sysdeps/sparc contents to sysdeps/sparc/sparc32, rename sparc8 subdir to sparcv8, move sysdeps/sparc64 content to sysdeps/sparc/sparc64. Same for sysdeps/unix/sysv/linux/sparc. * sysdeps/sparc/Implies: Do ieee754 here. * sysdeps/sparc/configure.in: Move from sparc32/. * sysdeps/sparc/sparc32/Implies: Do wordsize here. * sysdeps/sparc/sparc64/Implies: Likewise. * sysdeps/sparc/sparc32/dl-machine.h (elf_machine_load_address): Use uninitialized .got to find load address instead of AT_BASE, as the later does not exist when running ld.so directly. * sysdeps/sparc/sparc32/fpu/fpu_control.h: Move from ../. * sysdeps/sparc/sparc64/fpu/fpu_control.h: Likewise. * sysdeps/sparc/sparc32/udiv_qrnnd.S: Clean up PIC support. * sysdeps/sparc/sparc64/Makefile (sysdep_routines): Add s_finitel and s_signbitl for stdio. * sysdeps/sparc/sparc64/configure.in: Remove, we no longer need nopic_initfini. * sysdeps/sparc/sparc64/bsd-_setjmp.S: Remove. * sysdeps/sparc/sparc64/bsd-setjmp.S: Remove. * sysdeps/unix/sysv/linux/sparc/sparc64/bsd-_setjmp.S: New empty file. * sysdeps/unix/sysv/linux/sparc/sparc64/bsd-setjmp.S: New empty file. * sysdeps/unix/sysv/linux/sparc/sparc64/longjmp.S: Add __longjmp alias. * sysdeps/unix/sysv/linux/sparc/sparc64/setjmp.S: Add weak _setjmp and setjmp entry points. * sysdeps/sparc/sparc64/dl-machine.h: Finish real implementation. * sysdeps/sparc/sparc64/elf/crtbegin.S: Add copyright, fix PIC. * sysdeps/sparc/sparc64/elf/crtend.S: Likewise. * sysdeps/unix/sysv/linux/sparc/bits/sigaction.h (struct sigaction): Fix sa_flags size for sparc64; add sa_restorer for kernel. * sysdeps/unix/sysv/linux/sparc/sparc64/brk.S: Fix PIC. * sysdeps/unix/sysv/linux/sparc/sparc64/clone.S: Likewise. * sysdeps/unix/sysv/linux/sparc/sparc64/init-first.h: Likewise. * sysdeps/unix/sysv/linux/sparc/sparc64/sysdep.h: Likewise. * sysdeps/unix/sysv/linux/sparc/sparc64/syscall.S: New file. 1997-08-10 18:29:08 1997 Mark Kettenis * login/login.c (login): Check for correct return value of utmpname. 1997-08-11 16:49 Ulrich Drepper * time/tzset.c (__tzset_internal): Correctly handle TZ strings following incorrect old POSIX specs. Patch by Paul Eggert . * sysdeps/generic/bits/sigaction.h: Correct comment for SA_RESTART. * sysdeps/unix/bsd/osf/bits/sigaction.h: Likewise. * sysdeps/unix/sysv/linux/alpha/bits/sigaction.h: Likewise. * sysdeps/unix/sysv/linux/bits/sigaction.h: Likewise. * sysdeps/unix/sysv/linux/mips/bits/sigaction.h: Likewise. * sysdeps/unix/sysv/linux/sparc/bits/sigaction.h: Likewise. * sysdeps/unix/sysv/minix/bits/sigaction.h: Likewise. * sysdeps/unix/sysv/sysv4/bits/sigaction.h: Likewise. Reported by NIIBE Yutaka . 1997-08-09 Andreas Schwab * manual/math.texi: Define the macros @mul and @infinity and use them to format the multiplication and infinity signs. (FP Comparison Functions): Use @code, not @math, for C code examples. * elf/dl-profile.c: Add descriptions and implement reading of --- sysdeps/sparc/Dist | 4 - sysdeps/sparc/Implies | 1 - sysdeps/sparc/Makefile | 51 ---- sysdeps/sparc/__longjmp.S | 59 ----- sysdeps/sparc/add_n.S | 226 ------------------ sysdeps/sparc/addmul_1.S | 147 ------------ sysdeps/sparc/alloca.S | 32 --- sysdeps/sparc/bits/endian.h | 3 - sysdeps/sparc/bits/setjmp.h | 16 -- sysdeps/sparc/bsd-_setjmp.S | 40 ---- sysdeps/sparc/bsd-setjmp.S | 40 ---- sysdeps/sparc/divrem.m4 | 238 ------------------- sysdeps/sparc/dl-machine.h | 347 --------------------------- sysdeps/sparc/dotmul.S | 123 ---------- sysdeps/sparc/e_sqrt.c | 34 --- sysdeps/sparc/elf/Makefile | 4 - sysdeps/sparc/elf/start.c | 68 ------ sysdeps/sparc/fpu/bits/mathdef.h | 64 +++++ sysdeps/sparc/fpu/fclrexcpt.c | 32 +++ sysdeps/sparc/fpu/fegetenv.c | 26 ++ sysdeps/sparc/fpu/fegetround.c | 30 +++ sysdeps/sparc/fpu/feholdexcpt.c | 35 +++ sysdeps/sparc/fpu/fesetenv.c | 43 ++++ sysdeps/sparc/fpu/fesetround.c | 37 +++ sysdeps/sparc/fpu/feupdateenv.c | 38 +++ sysdeps/sparc/fpu/fgetexcptflg.c | 31 +++ sysdeps/sparc/fpu/fraiseexcpt.c | 66 ++++++ sysdeps/sparc/fpu/fsetexcptflg.c | 34 +++ sysdeps/sparc/fpu/ftestexcept.c | 30 +++ sysdeps/sparc/fpu_control.h | 69 ------ sysdeps/sparc/lshift.S | 95 -------- sysdeps/sparc/memcopy.h | 21 -- sysdeps/sparc/mul_1.S | 199 ---------------- sysdeps/sparc/rem.S | 369 ----------------------------- sysdeps/sparc/rshift.S | 92 ------- sysdeps/sparc/sdiv.S | 369 ----------------------------- sysdeps/sparc/setjmp.S | 54 ----- sysdeps/sparc/sparc32/Dist | 4 + sysdeps/sparc/sparc32/Implies | 1 + sysdeps/sparc/sparc32/Makefile | 51 ++++ sysdeps/sparc/sparc32/__longjmp.S | 59 +++++ sysdeps/sparc/sparc32/add_n.S | 226 ++++++++++++++++++ sysdeps/sparc/sparc32/addmul_1.S | 147 ++++++++++++ sysdeps/sparc/sparc32/alloca.S | 32 +++ sysdeps/sparc/sparc32/bits/endian.h | 3 + sysdeps/sparc/sparc32/bits/setjmp.h | 16 ++ sysdeps/sparc/sparc32/bsd-_setjmp.S | 40 ++++ sysdeps/sparc/sparc32/bsd-setjmp.S | 40 ++++ sysdeps/sparc/sparc32/divrem.m4 | 238 +++++++++++++++++++ sysdeps/sparc/sparc32/dl-machine.h | 363 ++++++++++++++++++++++++++++ sysdeps/sparc/sparc32/dotmul.S | 123 ++++++++++ sysdeps/sparc/sparc32/e_sqrt.c | 34 +++ sysdeps/sparc/sparc32/elf/Makefile | 4 + sysdeps/sparc/sparc32/elf/start.c | 68 ++++++ sysdeps/sparc/sparc32/fpu/bits/fenv.h | 76 ++++++ sysdeps/sparc/sparc32/fpu/fpu_control.h | 69 ++++++ sysdeps/sparc/sparc32/lshift.S | 95 ++++++++ sysdeps/sparc/sparc32/memcopy.h | 21 ++ sysdeps/sparc/sparc32/mul_1.S | 199 ++++++++++++++++ sysdeps/sparc/sparc32/rem.S | 369 +++++++++++++++++++++++++++++ sysdeps/sparc/sparc32/rshift.S | 92 +++++++ sysdeps/sparc/sparc32/sdiv.S | 369 +++++++++++++++++++++++++++++ sysdeps/sparc/sparc32/setjmp.S | 54 +++++ sysdeps/sparc/sparc32/sparcv8/addmul_1.S | 124 ++++++++++ sysdeps/sparc/sparc32/sparcv8/mul_1.S | 99 ++++++++ sysdeps/sparc/sparc32/sparcv8/submul_1.S | 58 +++++ sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S | 186 +++++++++++++++ sysdeps/sparc/sparc32/sub_n.S | 311 ++++++++++++++++++++++++ sysdeps/sparc/sparc32/submul_1.S | 147 ++++++++++++ sysdeps/sparc/sparc32/udiv.S | 352 +++++++++++++++++++++++++++ sysdeps/sparc/sparc32/udiv_qrnnd.S | 165 +++++++++++++ sysdeps/sparc/sparc32/umul.S | 153 ++++++++++++ sysdeps/sparc/sparc32/urem.S | 352 +++++++++++++++++++++++++++ sysdeps/sparc/sparc64/Implies | 1 + sysdeps/sparc/sparc64/Makefile | 11 + sysdeps/sparc/sparc64/add_n.S | 58 +++++ sysdeps/sparc/sparc64/addmul_1.S | 88 +++++++ sysdeps/sparc/sparc64/bits/endian.h | 8 + sysdeps/sparc/sparc64/dl-machine.h | 353 +++++++++++++++++++++++++++ sysdeps/sparc/sparc64/elf/Dist | 4 + sysdeps/sparc/sparc64/elf/Makefile | 10 + sysdeps/sparc/sparc64/elf/crtbegin.S | 69 ++++++ sysdeps/sparc/sparc64/elf/crtbeginS.S | 1 + sysdeps/sparc/sparc64/elf/crtend.S | 70 ++++++ sysdeps/sparc/sparc64/elf/crtendS.S | 1 + sysdeps/sparc/sparc64/elf/start.S | 93 ++++++++ sysdeps/sparc/sparc64/fpu/bits/fenv.h | 76 ++++++ sysdeps/sparc/sparc64/fpu/fpu_control.h | 69 ++++++ sysdeps/sparc/sparc64/gmp-mparam.h | 27 +++ sysdeps/sparc/sparc64/lshift.S | 96 ++++++++ sysdeps/sparc/sparc64/mul_1.S | 84 +++++++ sysdeps/sparc/sparc64/rshift.S | 93 ++++++++ sysdeps/sparc/sparc64/sub_n.S | 55 +++++ sysdeps/sparc/sparc64/submul_1.S | 87 +++++++ sysdeps/sparc/sparc8/addmul_1.S | 124 ---------- sysdeps/sparc/sparc8/mul_1.S | 99 -------- sysdeps/sparc/sparc8/submul_1.S | 58 ----- sysdeps/sparc/sparc8/udiv_qrnnd.S | 186 --------------- sysdeps/sparc/sub_n.S | 311 ------------------------ sysdeps/sparc/submul_1.S | 147 ------------ sysdeps/sparc/udiv.S | 352 --------------------------- sysdeps/sparc/udiv_qrnnd.S | 164 ------------- sysdeps/sparc/umul.S | 153 ------------ sysdeps/sparc/urem.S | 352 --------------------------- 104 files changed, 6560 insertions(+), 4647 deletions(-) delete mode 100644 sysdeps/sparc/Dist delete mode 100644 sysdeps/sparc/Makefile delete mode 100644 sysdeps/sparc/__longjmp.S delete mode 100644 sysdeps/sparc/add_n.S delete mode 100644 sysdeps/sparc/addmul_1.S delete mode 100644 sysdeps/sparc/alloca.S delete mode 100644 sysdeps/sparc/bits/endian.h delete mode 100644 sysdeps/sparc/bits/setjmp.h delete mode 100644 sysdeps/sparc/bsd-_setjmp.S delete mode 100644 sysdeps/sparc/bsd-setjmp.S delete mode 100644 sysdeps/sparc/divrem.m4 delete mode 100644 sysdeps/sparc/dl-machine.h delete mode 100644 sysdeps/sparc/dotmul.S delete mode 100644 sysdeps/sparc/e_sqrt.c delete mode 100644 sysdeps/sparc/elf/Makefile delete mode 100644 sysdeps/sparc/elf/start.c create mode 100644 sysdeps/sparc/fpu/bits/mathdef.h create mode 100644 sysdeps/sparc/fpu/fclrexcpt.c create mode 100644 sysdeps/sparc/fpu/fegetenv.c create mode 100644 sysdeps/sparc/fpu/fegetround.c create mode 100644 sysdeps/sparc/fpu/feholdexcpt.c create mode 100644 sysdeps/sparc/fpu/fesetenv.c create mode 100644 sysdeps/sparc/fpu/fesetround.c create mode 100644 sysdeps/sparc/fpu/feupdateenv.c create mode 100644 sysdeps/sparc/fpu/fgetexcptflg.c create mode 100644 sysdeps/sparc/fpu/fraiseexcpt.c create mode 100644 sysdeps/sparc/fpu/fsetexcptflg.c create mode 100644 sysdeps/sparc/fpu/ftestexcept.c delete mode 100644 sysdeps/sparc/fpu_control.h delete mode 100644 sysdeps/sparc/lshift.S delete mode 100644 sysdeps/sparc/memcopy.h delete mode 100644 sysdeps/sparc/mul_1.S delete mode 100644 sysdeps/sparc/rem.S delete mode 100644 sysdeps/sparc/rshift.S delete mode 100644 sysdeps/sparc/sdiv.S delete mode 100644 sysdeps/sparc/setjmp.S create mode 100644 sysdeps/sparc/sparc32/Dist create mode 100644 sysdeps/sparc/sparc32/Implies create mode 100644 sysdeps/sparc/sparc32/Makefile create mode 100644 sysdeps/sparc/sparc32/__longjmp.S create mode 100644 sysdeps/sparc/sparc32/add_n.S create mode 100644 sysdeps/sparc/sparc32/addmul_1.S create mode 100644 sysdeps/sparc/sparc32/alloca.S create mode 100644 sysdeps/sparc/sparc32/bits/endian.h create mode 100644 sysdeps/sparc/sparc32/bits/setjmp.h create mode 100644 sysdeps/sparc/sparc32/bsd-_setjmp.S create mode 100644 sysdeps/sparc/sparc32/bsd-setjmp.S create mode 100644 sysdeps/sparc/sparc32/divrem.m4 create mode 100644 sysdeps/sparc/sparc32/dl-machine.h create mode 100644 sysdeps/sparc/sparc32/dotmul.S create mode 100644 sysdeps/sparc/sparc32/e_sqrt.c create mode 100644 sysdeps/sparc/sparc32/elf/Makefile create mode 100644 sysdeps/sparc/sparc32/elf/start.c create mode 100644 sysdeps/sparc/sparc32/fpu/bits/fenv.h create mode 100644 sysdeps/sparc/sparc32/fpu/fpu_control.h create mode 100644 sysdeps/sparc/sparc32/lshift.S create mode 100644 sysdeps/sparc/sparc32/memcopy.h create mode 100644 sysdeps/sparc/sparc32/mul_1.S create mode 100644 sysdeps/sparc/sparc32/rem.S create mode 100644 sysdeps/sparc/sparc32/rshift.S create mode 100644 sysdeps/sparc/sparc32/sdiv.S create mode 100644 sysdeps/sparc/sparc32/setjmp.S create mode 100644 sysdeps/sparc/sparc32/sparcv8/addmul_1.S create mode 100644 sysdeps/sparc/sparc32/sparcv8/mul_1.S create mode 100644 sysdeps/sparc/sparc32/sparcv8/submul_1.S create mode 100644 sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S create mode 100644 sysdeps/sparc/sparc32/sub_n.S create mode 100644 sysdeps/sparc/sparc32/submul_1.S create mode 100644 sysdeps/sparc/sparc32/udiv.S create mode 100644 sysdeps/sparc/sparc32/udiv_qrnnd.S create mode 100644 sysdeps/sparc/sparc32/umul.S create mode 100644 sysdeps/sparc/sparc32/urem.S create mode 100644 sysdeps/sparc/sparc64/Implies create mode 100644 sysdeps/sparc/sparc64/Makefile create mode 100644 sysdeps/sparc/sparc64/add_n.S create mode 100644 sysdeps/sparc/sparc64/addmul_1.S create mode 100644 sysdeps/sparc/sparc64/bits/endian.h create mode 100644 sysdeps/sparc/sparc64/dl-machine.h create mode 100644 sysdeps/sparc/sparc64/elf/Dist create mode 100644 sysdeps/sparc/sparc64/elf/Makefile create mode 100644 sysdeps/sparc/sparc64/elf/crtbegin.S create mode 100644 sysdeps/sparc/sparc64/elf/crtbeginS.S create mode 100644 sysdeps/sparc/sparc64/elf/crtend.S create mode 100644 sysdeps/sparc/sparc64/elf/crtendS.S create mode 100644 sysdeps/sparc/sparc64/elf/start.S create mode 100644 sysdeps/sparc/sparc64/fpu/bits/fenv.h create mode 100644 sysdeps/sparc/sparc64/fpu/fpu_control.h create mode 100644 sysdeps/sparc/sparc64/gmp-mparam.h create mode 100644 sysdeps/sparc/sparc64/lshift.S create mode 100644 sysdeps/sparc/sparc64/mul_1.S create mode 100644 sysdeps/sparc/sparc64/rshift.S create mode 100644 sysdeps/sparc/sparc64/sub_n.S create mode 100644 sysdeps/sparc/sparc64/submul_1.S delete mode 100644 sysdeps/sparc/sparc8/addmul_1.S delete mode 100644 sysdeps/sparc/sparc8/mul_1.S delete mode 100644 sysdeps/sparc/sparc8/submul_1.S delete mode 100644 sysdeps/sparc/sparc8/udiv_qrnnd.S delete mode 100644 sysdeps/sparc/sub_n.S delete mode 100644 sysdeps/sparc/submul_1.S delete mode 100644 sysdeps/sparc/udiv.S delete mode 100644 sysdeps/sparc/udiv_qrnnd.S delete mode 100644 sysdeps/sparc/umul.S delete mode 100644 sysdeps/sparc/urem.S (limited to 'sysdeps/sparc') diff --git a/sysdeps/sparc/Dist b/sysdeps/sparc/Dist deleted file mode 100644 index 6c3b23baa5..0000000000 --- a/sysdeps/sparc/Dist +++ /dev/null @@ -1,4 +0,0 @@ -dotmul.S umul.S -divrem.m4 sdiv.S udiv.S rem.S urem.S -alloca.S -sys/trap.h diff --git a/sysdeps/sparc/Implies b/sysdeps/sparc/Implies index 8a2007c5ce..da719e1707 100644 --- a/sysdeps/sparc/Implies +++ b/sysdeps/sparc/Implies @@ -1,3 +1,2 @@ -wordsize-32 # SPARC uses IEEE 754 floating point. ieee754 diff --git a/sysdeps/sparc/Makefile b/sysdeps/sparc/Makefile deleted file mode 100644 index deec2f8145..0000000000 --- a/sysdeps/sparc/Makefile +++ /dev/null @@ -1,51 +0,0 @@ -# Copyright (C) 1991, 92, 93, 94, 95, 96, 97 Free Software Foundation, Inc. -# This file is part of the GNU C Library. - -# The GNU C Library is free software; you can redistribute it and/or -# modify it under the terms of the GNU Library General Public License -# as published by the Free Software Foundation; either version 2 of -# the License, or (at your option) any later version. - -# The GNU C Library is distributed in the hope that it will be useful, -# but WITHOUT ANY WARRANTY; without even the implied warranty of -# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU -# Library General Public License for more details. - -# You should have received a copy of the GNU Library General Public -# License along with the GNU C Library; see the file COPYING.LIB. If not, -# write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, -# Boston, MA 02111-1307, USA. - -ifeq ($(subdir),gnulib) -routines = dotmul umul $(divrem) alloca -endif # gnulib - -# We distribute these files, even though they are generated, -# so as to avoid the need for a functioning m4 to build the library. -divrem := sdiv udiv rem urem - -+divrem-NAME-sdiv := div -+divrem-NAME-udiv := udiv -+divrem-NAME-rem := rem -+divrem-NAME-urem := urem -+divrem-NAME = $(+divrem-NAME-$(basename $(notdir $@))) -+divrem-OP-div := div -+divrem-OP-udiv := div -+divrem-OP-rem := rem -+divrem-OP-urem := rem -+divrem-S-div := true -+divrem-S-rem := true -+divrem-S-udiv := false -+divrem-S-urem := false -$(divrem:%=$(sysdep_dir)/sparc/%.S): $(sysdep_dir)/sparc/divrem.m4 - (echo "define(NAME,\`.$(+divrem-NAME)')\ - define(OP,\`$(+divrem-OP-$(+divrem-NAME))')\ - define(S,\`$(+divrem-S-$(+divrem-NAME))')\ - /* This file is generated from divrem.m4; DO NOT EDIT! */"; \ - cat $<) | $(M4) > $@-tmp -# Make it unwritable so noone will edit it by mistake. - -chmod a-w $@-tmp - mv -f $@-tmp $@ - test ! -d CVS || cvs commit -m'Regenerated from $<' $@ - -sysdep-realclean := $(sysdep-realclean) $(divrem:%=sysdeps/sparc/%.S) diff --git a/sysdeps/sparc/__longjmp.S b/sysdeps/sparc/__longjmp.S deleted file mode 100644 index 36e1c170dd..0000000000 --- a/sysdeps/sparc/__longjmp.S +++ /dev/null @@ -1,59 +0,0 @@ -/* Copyright (C) 1991, 1993, 1996, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include - -#define _ASM 1 -#include -#define ENV(reg) [%g1 + (reg * 4)] - -ENTRY (__longjmp) - /* Store our arguments in global registers so we can still - use them while unwinding frames and their register windows. */ - mov %o0, %g1 /* ENV in %g1 */ - orcc %o1, %g0, %g6 /* VAL in %g6 */ - be,a 0f /* Branch if zero; else skip delay slot. */ - mov 1, %g6 /* Delay slot only hit if zero: VAL = 1. */ -0: - - /* Cache target FP in register %g7. */ - ld ENV (JB_FP), %g7 - - /* Now we will loop, unwinding the register windows up the stack - until the restored %fp value matches the target value in %g7. */ - -loop: cmp %fp, %g7 /* Have we reached the target frame? */ - bl,a loop /* Loop while current fp is below target. */ - restore /* Unwind register window in delay slot. */ - be,a found /* Better have hit it exactly. */ - ld ENV (JB_SP), %o0 /* Delay slot: extract target SP. */ - -bogus: /* Get here only if the jmp_buf or stack is clobbered. */ - call C_SYMBOL_NAME (abort) - nop - unimp 0 - -found: /* We have unwound register windows so %fp matches the target. */ - cmp %o0, %sp /* Check jmp_buf SP vs register window. */ - bge,a sp_ok /* Saved must not be deeper than register. */ - mov %o0, %sp /* OK, install new SP. */ - b,a bogus /* Bogus, we lose. */ - -sp_ok: ld ENV (JB_PC), %o0 /* Extract target return PC. */ - jmp %o0 + 8 /* Return there. */ - mov %g6, %o0 /* Delay slot: set return value. */ diff --git a/sysdeps/sparc/add_n.S b/sysdeps/sparc/add_n.S deleted file mode 100644 index 9852c256aa..0000000000 --- a/sysdeps/sparc/add_n.S +++ /dev/null @@ -1,226 +0,0 @@ -! SPARC __mpn_add_n -- Add two limb vectors of the same length > 0 and store -! sum in a third limb vector. - -! Copyright (C) 1995, 1996 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -#define res_ptr %o0 -#define s1_ptr %o1 -#define s2_ptr %o2 -#define size %o3 - -#include "sysdep.h" - - .text - .align 4 - .global C_SYMBOL_NAME(__mpn_add_n) -C_SYMBOL_NAME(__mpn_add_n): - xor s2_ptr,res_ptr,%g1 - andcc %g1,4,%g0 - bne L1 ! branch if alignment differs - nop -! ** V1a ** -L0: andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 - be L_v1 ! if no, branch - nop -/* Add least significant limb separately to align res_ptr and s2_ptr */ - ld [s1_ptr],%g4 - add s1_ptr,4,s1_ptr - ld [s2_ptr],%g2 - add s2_ptr,4,s2_ptr - add size,-1,size - addcc %g4,%g2,%o4 - st %o4,[res_ptr] - add res_ptr,4,res_ptr -L_v1: addx %g0,%g0,%o4 ! save cy in register - cmp size,2 ! if size < 2 ... - bl Lend2 ! ... branch to tail code - subcc %g0,%o4,%g0 ! restore cy - - ld [s1_ptr+0],%g4 - addcc size,-10,size - ld [s1_ptr+4],%g1 - ldd [s2_ptr+0],%g2 - blt Lfin1 - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 8 limbs until less than 8 limbs remain */ -Loop1: addxcc %g4,%g2,%o4 - ld [s1_ptr+8],%g4 - addxcc %g1,%g3,%o5 - ld [s1_ptr+12],%g1 - ldd [s2_ptr+8],%g2 - std %o4,[res_ptr+0] - addxcc %g4,%g2,%o4 - ld [s1_ptr+16],%g4 - addxcc %g1,%g3,%o5 - ld [s1_ptr+20],%g1 - ldd [s2_ptr+16],%g2 - std %o4,[res_ptr+8] - addxcc %g4,%g2,%o4 - ld [s1_ptr+24],%g4 - addxcc %g1,%g3,%o5 - ld [s1_ptr+28],%g1 - ldd [s2_ptr+24],%g2 - std %o4,[res_ptr+16] - addxcc %g4,%g2,%o4 - ld [s1_ptr+32],%g4 - addxcc %g1,%g3,%o5 - ld [s1_ptr+36],%g1 - ldd [s2_ptr+32],%g2 - std %o4,[res_ptr+24] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - add s1_ptr,32,s1_ptr - add s2_ptr,32,s2_ptr - add res_ptr,32,res_ptr - bge Loop1 - subcc %g0,%o4,%g0 ! restore cy - -Lfin1: addcc size,8-2,size - blt Lend1 - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 2 limbs until less than 2 limbs remain */ -Loope1: addxcc %g4,%g2,%o4 - ld [s1_ptr+8],%g4 - addxcc %g1,%g3,%o5 - ld [s1_ptr+12],%g1 - ldd [s2_ptr+8],%g2 - std %o4,[res_ptr+0] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-2,size - add s1_ptr,8,s1_ptr - add s2_ptr,8,s2_ptr - add res_ptr,8,res_ptr - bge Loope1 - subcc %g0,%o4,%g0 ! restore cy -Lend1: addxcc %g4,%g2,%o4 - addxcc %g1,%g3,%o5 - std %o4,[res_ptr+0] - addx %g0,%g0,%o4 ! save cy in register - - andcc size,1,%g0 - be Lret1 - subcc %g0,%o4,%g0 ! restore cy -/* Add last limb */ - ld [s1_ptr+8],%g4 - ld [s2_ptr+8],%g2 - addxcc %g4,%g2,%o4 - st %o4,[res_ptr+8] - -Lret1: retl - addx %g0,%g0,%o0 ! return carry-out from most sign. limb - -L1: xor s1_ptr,res_ptr,%g1 - andcc %g1,4,%g0 - bne L2 - nop -! ** V1b ** - mov s2_ptr,%g1 - mov s1_ptr,s2_ptr - b L0 - mov %g1,s1_ptr - -! ** V2 ** -/* If we come here, the alignment of s1_ptr and res_ptr as well as the - alignment of s2_ptr and res_ptr differ. Since there are only two ways - things can be aligned (that we care about) we now know that the alignment - of s1_ptr and s2_ptr are the same. */ - -L2: cmp size,1 - be Ljone - nop - andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 - be L_v2 ! if no, branch - nop -/* Add least significant limb separately to align s1_ptr and s2_ptr */ - ld [s1_ptr],%g4 - add s1_ptr,4,s1_ptr - ld [s2_ptr],%g2 - add s2_ptr,4,s2_ptr - add size,-1,size - addcc %g4,%g2,%o4 - st %o4,[res_ptr] - add res_ptr,4,res_ptr - -L_v2: addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - blt Lfin2 - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 8 limbs until less than 8 limbs remain */ -Loop2: ldd [s1_ptr+0],%g2 - ldd [s2_ptr+0],%o4 - addxcc %g2,%o4,%g2 - st %g2,[res_ptr+0] - addxcc %g3,%o5,%g3 - st %g3,[res_ptr+4] - ldd [s1_ptr+8],%g2 - ldd [s2_ptr+8],%o4 - addxcc %g2,%o4,%g2 - st %g2,[res_ptr+8] - addxcc %g3,%o5,%g3 - st %g3,[res_ptr+12] - ldd [s1_ptr+16],%g2 - ldd [s2_ptr+16],%o4 - addxcc %g2,%o4,%g2 - st %g2,[res_ptr+16] - addxcc %g3,%o5,%g3 - st %g3,[res_ptr+20] - ldd [s1_ptr+24],%g2 - ldd [s2_ptr+24],%o4 - addxcc %g2,%o4,%g2 - st %g2,[res_ptr+24] - addxcc %g3,%o5,%g3 - st %g3,[res_ptr+28] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - add s1_ptr,32,s1_ptr - add s2_ptr,32,s2_ptr - add res_ptr,32,res_ptr - bge Loop2 - subcc %g0,%o4,%g0 ! restore cy - -Lfin2: addcc size,8-2,size - blt Lend2 - subcc %g0,%o4,%g0 ! restore cy -Loope2: ldd [s1_ptr+0],%g2 - ldd [s2_ptr+0],%o4 - addxcc %g2,%o4,%g2 - st %g2,[res_ptr+0] - addxcc %g3,%o5,%g3 - st %g3,[res_ptr+4] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-2,size - add s1_ptr,8,s1_ptr - add s2_ptr,8,s2_ptr - add res_ptr,8,res_ptr - bge Loope2 - subcc %g0,%o4,%g0 ! restore cy -Lend2: andcc size,1,%g0 - be Lret2 - subcc %g0,%o4,%g0 ! restore cy -/* Add last limb */ -Ljone: ld [s1_ptr],%g4 - ld [s2_ptr],%g2 - addxcc %g4,%g2,%o4 - st %o4,[res_ptr] - -Lret2: retl - addx %g0,%g0,%o0 ! return carry-out from most sign. limb diff --git a/sysdeps/sparc/addmul_1.S b/sysdeps/sparc/addmul_1.S deleted file mode 100644 index 375d25db6b..0000000000 --- a/sysdeps/sparc/addmul_1.S +++ /dev/null @@ -1,147 +0,0 @@ -! SPARC __mpn_addmul_1 -- Multiply a limb vector with a limb and add -! the result to a second limb vector. - -! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr o0 -! s1_ptr o1 -! size o2 -! s2_limb o3 - -#include "sysdep.h" - -.text - .align 4 - .global C_SYMBOL_NAME(__mpn_addmul_1) -C_SYMBOL_NAME(__mpn_addmul_1): - ! Make S1_PTR and RES_PTR point at the end of their blocks - ! and put (- 4 x SIZE) in index/loop counter. - sll %o2,2,%o2 - add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval - add %o1,%o2,%o1 - sub %g0,%o2,%o2 - - cmp %o3,0xfff - bgu Large - nop - - ld [%o1+%o2],%o5 - mov 0,%o0 - b L0 - add %o4,-4,%o4 -Loop0: - addcc %o5,%g1,%g1 - ld [%o1+%o2],%o5 - addx %o0,%g0,%o0 - st %g1,[%o4+%o2] -L0: wr %g0,%o3,%y - sra %o5,31,%g2 - and %o3,%g2,%g2 - andcc %g1,0,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,0,%g1 - sra %g1,20,%g4 - sll %g1,12,%g1 - rd %y,%g3 - srl %g3,20,%g3 - or %g1,%g3,%g1 - - addcc %g1,%o0,%g1 - addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb - addcc %o2,4,%o2 ! loop counter - bne Loop0 - ld [%o4+%o2],%o5 - - addcc %o5,%g1,%g1 - addx %o0,%g0,%o0 - retl - st %g1,[%o4+%o2] - - -Large: ld [%o1+%o2],%o5 - mov 0,%o0 - sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 - b L1 - add %o4,-4,%o4 -Loop: - addcc %o5,%g3,%g3 - ld [%o1+%o2],%o5 - addx %o0,%g0,%o0 - st %g3,[%o4+%o2] -L1: wr %g0,%o5,%y - and %o5,%g4,%g2 - andcc %g0,%g0,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%g0,%g1 - rd %y,%g3 - addcc %g3,%o0,%g3 - addx %g2,%g1,%o0 - addcc %o2,4,%o2 - bne Loop - ld [%o4+%o2],%o5 - - addcc %o5,%g3,%g3 - addx %o0,%g0,%o0 - retl - st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/alloca.S b/sysdeps/sparc/alloca.S deleted file mode 100644 index dcbd171163..0000000000 --- a/sysdeps/sparc/alloca.S +++ /dev/null @@ -1,32 +0,0 @@ -/* Copyright (C) 1994, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include "sysdep.h" - -/* Code produced by Sun's C compiler calls this function with two extra - arguments which it makes relocatable symbols but seem always to be - the constant 96; I have no idea what they are for. */ - -#ifndef NO_UNDERSCORES -#define __builtin_alloca ___builtin_alloca -#endif - -ENTRY (__builtin_alloca) - sub %sp, %o0, %sp /* Push some stack space. */ - retl /* Return; the returned buffer leaves 96 */ - add %sp, 96, %o0 /* bytes of register save area at the top. */ diff --git a/sysdeps/sparc/bits/endian.h b/sysdeps/sparc/bits/endian.h deleted file mode 100644 index f1a75c0652..0000000000 --- a/sysdeps/sparc/bits/endian.h +++ /dev/null @@ -1,3 +0,0 @@ -/* SPARC is big-endian. */ - -#define __BYTE_ORDER __BIG_ENDIAN diff --git a/sysdeps/sparc/bits/setjmp.h b/sysdeps/sparc/bits/setjmp.h deleted file mode 100644 index 43bae1a496..0000000000 --- a/sysdeps/sparc/bits/setjmp.h +++ /dev/null @@ -1,16 +0,0 @@ -/* Define the machine-dependent type `jmp_buf'. SPARC version. */ - -#if defined (__USE_MISC) || defined (_ASM) -#define JB_SP 0 -#define JB_FP 1 -#define JB_PC 2 -#endif - -#ifndef _ASM -typedef int __jmp_buf[3]; -#endif - -/* Test if longjmp to JMPBUF would unwind the frame - containing a local variable at ADDRESS. */ -#define _JMPBUF_UNWINDS(jmpbuf, address) \ - ((int) (address) < (jmpbuf)[JB_SP]) diff --git a/sysdeps/sparc/bsd-_setjmp.S b/sysdeps/sparc/bsd-_setjmp.S deleted file mode 100644 index bf49765f4d..0000000000 --- a/sysdeps/sparc/bsd-_setjmp.S +++ /dev/null @@ -1,40 +0,0 @@ -/* BSD `_setjmp' entry point to `sigsetjmp (..., 0)'. Sparc version. - Copyright (C) 1994, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include - -ENTRY (_setjmp) - -#ifdef PIC - save %sp, -64, %sp -1: call 2f - sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 -2: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 - add %g1, %o7, %g1 - sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g2 - restore - or %g2, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g2 - ld [%g1+%g2], %g1 -#else - sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g1 - or %g1, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g1 -#endif - - jmp %g1 - mov %g0, %o1 /* Pass second argument of zero. */ diff --git a/sysdeps/sparc/bsd-setjmp.S b/sysdeps/sparc/bsd-setjmp.S deleted file mode 100644 index 16dc260765..0000000000 --- a/sysdeps/sparc/bsd-setjmp.S +++ /dev/null @@ -1,40 +0,0 @@ -/* BSD `setjmp' entry point to `sigsetjmp (..., 1)'. Sparc version. - Copyright (C) 1994, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include - -ENTRY (setjmp) - -#ifdef PIC - save %sp, -64, %sp -1: call 2f - sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 -2: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 - add %g1, %o7, %g1 - sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g2 - restore - or %g2, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g2 - ld [%g1+%g2], %g1 -#else - sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g1 - or %g1, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g1 -#endif - - jmp %g1 - mov 1, %o1 /* Pass second argument of one. */ diff --git a/sysdeps/sparc/divrem.m4 b/sysdeps/sparc/divrem.m4 deleted file mode 100644 index 665abf11ae..0000000000 --- a/sysdeps/sparc/divrem.m4 +++ /dev/null @@ -1,238 +0,0 @@ -/* - * Division and remainder, from Appendix E of the Sparc Version 8 - * Architecture Manual, with fixes from Gordon Irlam. - */ - -/* - * Input: dividend and divisor in %o0 and %o1 respectively. - * - * m4 parameters: - * NAME name of function to generate - * OP OP=div => %o0 / %o1; OP=rem => %o0 % %o1 - * S S=true => signed; S=false => unsigned - * - * Algorithm parameters: - * N how many bits per iteration we try to get (4) - * WORDSIZE total number of bits (32) - * - * Derived constants: - * TOPBITS number of bits in the top `decade' of a number - * - * Important variables: - * Q the partial quotient under development (initially 0) - * R the remainder so far, initially the dividend - * ITER number of main division loop iterations required; - * equal to ceil(log2(quotient) / N). Note that this - * is the log base (2^N) of the quotient. - * V the current comparand, initially divisor*2^(ITER*N-1) - * - * Cost: - * Current estimate for non-large dividend is - * ceil(log2(quotient) / N) * (10 + 7N/2) + C - * A large dividend is one greater than 2^(31-TOPBITS) and takes a - * different path, as the upper bits of the quotient must be developed - * one bit at a time. - */ - -define(N, `4')dnl -define(WORDSIZE, `32')dnl -define(TOPBITS, eval(WORDSIZE - N*((WORDSIZE-1)/N)))dnl -dnl -define(dividend, `%o0')dnl -define(divisor, `%o1')dnl -define(Q, `%o2')dnl -define(R, `%o3')dnl -define(ITER, `%o4')dnl -define(V, `%o5')dnl -dnl -dnl m4 reminder: ifelse(a,b,c,d) => if a is b, then c, else d -define(T, `%g1')dnl -define(SC, `%g7')dnl -ifelse(S, `true', `define(SIGN, `%g6')')dnl - -dnl -dnl This is the recursive definition for developing quotient digits. -dnl -dnl Parameters: -dnl $1 the current depth, 1 <= $1 <= N -dnl $2 the current accumulation of quotient bits -dnl N max depth -dnl -dnl We add a new bit to $2 and either recurse or insert the bits in -dnl the quotient. R, Q, and V are inputs and outputs as defined above; -dnl the condition codes are expected to reflect the input R, and are -dnl modified to reflect the output R. -dnl -define(DEVELOP_QUOTIENT_BITS, -` ! depth $1, accumulated bits $2 - bl L.$1.eval(2**N+$2) - srl V,1,V - ! remainder is positive - subcc R,V,R - ifelse($1, N, - ` b 9f - add Q, ($2*2+1), Q - ', ` DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2+1)')') -L.$1.eval(2**N+$2): - ! remainder is negative - addcc R,V,R - ifelse($1, N, - ` b 9f - add Q, ($2*2-1), Q - ', ` DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2-1)')') - ifelse($1, 1, `9:')')dnl - -#include "sysdep.h" -#ifdef __linux__ -#include -#else -#ifdef __svr4__ -#include -#else -#include -#endif -#endif - -FUNC(NAME) -ifelse(S, `true', -` ! compute sign of result; if neither is negative, no problem - orcc divisor, dividend, %g0 ! either negative? - bge 2f ! no, go do the divide -ifelse(OP, `div', -` xor divisor, dividend, SIGN ! compute sign in any case', -` mov dividend, SIGN ! sign of remainder matches dividend') - tst divisor - bge 1f - tst dividend - ! divisor is definitely negative; dividend might also be negative - bge 2f ! if dividend not negative... - sub %g0, divisor, divisor ! in any case, make divisor nonneg -1: ! dividend is negative, divisor is nonnegative - sub %g0, dividend, dividend ! make dividend nonnegative -2: -') - ! Ready to divide. Compute size of quotient; scale comparand. - orcc divisor, %g0, V - bne 1f - mov dividend, R - - ! Divide by zero trap. If it returns, return 0 (about as - ! wrong as possible, but that is what SunOS does...). - ta ST_DIV0 - retl - clr %o0 - -1: - cmp R, V ! if divisor exceeds dividend, done - blu Lgot_result ! (and algorithm fails otherwise) - clr Q - sethi %hi(1 << (WORDSIZE - TOPBITS - 1)), T - cmp R, T - blu Lnot_really_big - clr ITER - - ! `Here the dividend is >= 2**(31-N) or so. We must be careful here, - ! as our usual N-at-a-shot divide step will cause overflow and havoc. - ! The number of bits in the result here is N*ITER+SC, where SC <= N. - ! Compute ITER in an unorthodox manner: know we need to shift V into - ! the top decade: so do not even bother to compare to R.' - 1: - cmp V, T - bgeu 3f - mov 1, SC - sll V, N, V - b 1b - add ITER, 1, ITER - - ! Now compute SC. - 2: addcc V, V, V - bcc Lnot_too_big - add SC, 1, SC - - ! We get here if the divisor overflowed while shifting. - ! This means that R has the high-order bit set. - ! Restore V and subtract from R. - sll T, TOPBITS, T ! high order bit - srl V, 1, V ! rest of V - add V, T, V - b Ldo_single_div - sub SC, 1, SC - - Lnot_too_big: - 3: cmp V, R - blu 2b - nop - be Ldo_single_div - nop - /* NB: these are commented out in the V8-Sparc manual as well */ - /* (I do not understand this) */ - ! V > R: went too far: back up 1 step - ! srl V, 1, V - ! dec SC - ! do single-bit divide steps - ! - ! We have to be careful here. We know that R >= V, so we can do the - ! first divide step without thinking. BUT, the others are conditional, - ! and are only done if R >= 0. Because both R and V may have the high- - ! order bit set in the first step, just falling into the regular - ! division loop will mess up the first time around. - ! So we unroll slightly... - Ldo_single_div: - subcc SC, 1, SC - bl Lend_regular_divide - nop - sub R, V, R - mov 1, Q - b Lend_single_divloop - nop - Lsingle_divloop: - sll Q, 1, Q - bl 1f - srl V, 1, V - ! R >= 0 - sub R, V, R - b 2f - add Q, 1, Q - 1: ! R < 0 - add R, V, R - sub Q, 1, Q - 2: - Lend_single_divloop: - subcc SC, 1, SC - bge Lsingle_divloop - tst R - b,a Lend_regular_divide - -Lnot_really_big: -1: - sll V, N, V - cmp V, R - bleu 1b - addcc ITER, 1, ITER - be Lgot_result - sub ITER, 1, ITER - - tst R ! set up for initial iteration -Ldivloop: - sll Q, N, Q - DEVELOP_QUOTIENT_BITS(1, 0) -Lend_regular_divide: - subcc ITER, 1, ITER - bge Ldivloop - tst R - bl,a Lgot_result - ! non-restoring fixup here (one instruction only!) -ifelse(OP, `div', -` sub Q, 1, Q -', ` add R, divisor, R -') - -Lgot_result: -ifelse(S, `true', -` ! check to see if answer should be < 0 - tst SIGN - bl,a 1f - ifelse(OP, `div', `sub %g0, Q, Q', `sub %g0, R, R') -1:') - retl - ifelse(OP, `div', `mov Q, %o0', `mov R, %o0') diff --git a/sysdeps/sparc/dl-machine.h b/sysdeps/sparc/dl-machine.h deleted file mode 100644 index dc38e17292..0000000000 --- a/sysdeps/sparc/dl-machine.h +++ /dev/null @@ -1,347 +0,0 @@ -/* Machine-dependent ELF dynamic relocation inline functions. SPARC version. - Copyright (C) 1996, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If - not, write to the Free Software Foundation, Inc., - 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ - -#define ELF_MACHINE_NAME "sparc" - -#include -#include -#include -#include - - -/* Some SPARC opcodes we need to use for self-modifying code. */ -#define OPCODE_NOP 0x01000000 /* nop */ -#define OPCODE_CALL 0x40000000 /* call ?; add PC-rel word address */ -#define OPCODE_SETHI_G1 0x03000000 /* sethi ?, %g1; add value>>10 */ -#define OPCODE_JMP_G1 0x81c06000 /* jmp %g1+?; add lo 10 bits of value */ -#define OPCODE_SAVE_SP64 0x9de3bfc0 /* save %sp, -64, %sp */ - - -/* Return nonzero iff E_MACHINE is compatible with the running host. */ -static inline int -elf_machine_matches_host (Elf32_Half e_machine) -{ - return e_machine == EM_SPARC; -} - - -/* Return the link-time address of _DYNAMIC. Conveniently, this is the - first element of the GOT. This must be inlined in a function which - uses global data. */ -static inline Elf32_Addr -elf_machine_dynamic (void) -{ - register Elf32_Addr *got asm ("%l7"); - return *got; -} - - -/* Return the run-time load address of the shared object. */ -static inline Elf32_Addr -elf_machine_load_address (void) -{ - Elf32_Addr addr; - - asm ( - "add %%fp,0x44,%%o2\n\t" /* o2 = point to argc */ - "ld [%%o2 - 4],%%o0\n\t" /* o0 = load argc */ - "sll %%o0, 2, %%o0\n\t" /* o0 = argc * sizeof (int) */ - "add %%o2,%%o0,%%o2\n\t" /* o2 = skip over argv */ - "add %%o2,4,%%o2\n\t" /* skip over null after argv */ - - /* Now %o2 is pointing to env, skip over that as well. */ - "1:\n\t" - "ld [%%o2],%%o0\n\t" - "cmp %%o0,0\n\t" - "bnz 1b\n\t" - "add %%o2,4,%%o2\n\t" - - /* Note that above, we want to advance the NULL after envp so - we always add 4. */ - - /* Now, search for the AT_BASE property. */ - "2:\n\t" - "ld [%%o2],%%o0\n\t" - "cmp %%o0,0\n\t" - "be,a 3f\n\t" - "or %%g0,%%g0,%0\n\t" - "cmp %%o0,7\n\t" /* AT_BASE = 7 */ - "be,a 3f\n\t" - "ld [%%o2+4],%0\n\t" - "b 2b\n\t" - "add %%o2,8,%%o2\n\t" - /* At this point %0 has the load address for the interpreter */ - "3:\n\t" - : "=r" (addr) - : /* no inputs */ - : "o0", "o2"); - return addr; -} - -#ifdef RESOLVE -/* Perform the relocation specified by RELOC and SYM (which is fully resolved). - MAP is the object containing the reloc. */ - -static inline void -elf_machine_rela (struct link_map *map, const Elf32_Rela *reloc, - const Elf32_Sym *sym, const struct r_found_version *version, - Elf32_Addr *const reloc_addr) -{ - Elf32_Addr loadbase; - - if (ELF32_R_TYPE (reloc->r_info) == R_SPARC_RELATIVE) - { -#ifndef RTLD_BOOTSTRAP - if (map != &_dl_rtld_map) /* Already done in rtld itself. */ -#endif - *reloc_addr += map->l_addr + reloc->r_addend; - } - else - { - const Elf32_Sym *const refsym = sym; - Elf32_Addr value; - if (sym->st_shndx != SHN_UNDEF && - ELF32_ST_BIND (sym->st_info) == STB_LOCAL) - value = map->l_addr; - else - { - value = RESOLVE (&sym, version, ELF32_R_TYPE (reloc->r_info)); - if (sym) - value += sym->st_value; - } - value += reloc->r_addend; /* Assume copy relocs have zero addend. */ - - switch (ELF32_R_TYPE (reloc->r_info)) - { - case R_SPARC_COPY: - if (sym->st_size > refsym->st_size - || (_dl_verbose && sym->st_size < refsym->st_size)) - { - extern char **_dl_argv; - const char *strtab; - - strtab = ((void *) map->l_addr - + map->l_info[DT_STRTAB]->d_un.d_ptr); - _dl_sysdep_error (_dl_argv[0] ?: "", - ": Symbol `", strtab + refsym->st_name, - "' has different size in shared object, " - "consider re-linking\n", NULL); - } - memcpy (reloc_addr, (void *) value, MIN (sym->st_size, - refsym->st_size)); - break; - case R_SPARC_GLOB_DAT: - case R_SPARC_32: - *reloc_addr = value; - break; - case R_SPARC_JMP_SLOT: - reloc_addr[1] = OPCODE_SETHI_G1 | (value >> 10); - reloc_addr[2] = OPCODE_JMP_G1 | (value & 0x3ff); - break; - case R_SPARC_8: - *(char *) reloc_addr = value; - break; - case R_SPARC_16: - *(short *) reloc_addr = value; - break; - case R_SPARC_DISP8: - *(char *) reloc_addr = (value - (Elf32_Addr) reloc_addr); - break; - case R_SPARC_DISP16: - *(short *) reloc_addr = (value - (Elf32_Addr) reloc_addr); - break; - case R_SPARC_DISP32: - *reloc_addr = (value - (Elf32_Addr) reloc_addr); - break; - case R_SPARC_LO10: - *reloc_addr = (*reloc_addr & ~0x3ff) | (value & 0x3ff); - break; - case R_SPARC_WDISP30: - *reloc_addr = ((*reloc_addr & 0xc0000000) - | ((value - (unsigned int) reloc_addr) >> 2)); - break; - case R_SPARC_HI22: - *reloc_addr = (*reloc_addr & 0xffc00000) | (value >> 10); - break; - case R_SPARC_NONE: /* Alright, Wilbur. */ - break; - default: - assert (! "unexpected dynamic reloc type"); - break; - } - } -} - -static inline void -elf_machine_lazy_rel (struct link_map *map, const Elf32_Rela *reloc) -{ - switch (ELF32_R_TYPE (reloc->r_info)) - { - case R_SPARC_NONE: - break; - case R_SPARC_JMP_SLOT: - break; - default: - assert (! "unexpected PLT reloc type"); - break; - } -} - -#endif /* RESOLVE */ - -/* Nonzero iff TYPE should not be allowed to resolve to one of - the main executable's symbols, as for a COPY reloc. */ -#define elf_machine_lookup_noexec_p(type) ((type) == R_SPARC_COPY) - -/* Nonzero iff TYPE describes relocation of a PLT entry, so - PLT entries should not be allowed to define the value. */ -#define elf_machine_lookup_noplt_p(type) ((type) == R_SPARC_JMP_SLOT) - -/* A reloc type used for ld.so cmdline arg lookups to reject PLT entries. */ -#define ELF_MACHINE_RELOC_NOPLT R_SPARC_JMP_SLOT - -/* The SPARC never uses Elf32_Rel relocations. */ -#define ELF_MACHINE_NO_REL 1 - - -/* Set up the loaded object described by L so its unrelocated PLT - entries will jump to the on-demand fixup code in dl-runtime.c. */ - -static inline int -elf_machine_runtime_setup (struct link_map *l, int lazy, int profile) -{ - Elf32_Addr *plt; - extern void _dl_runtime_resolve (Elf32_Word); - - if (l->l_info[DT_JMPREL] && lazy) - { - /* The entries for functions in the PLT have not yet been filled in. - Their initial contents will arrange when called to set the high 22 - bits of %g1 with an offset into the .rela.plt section and jump to - the beginning of the PLT. */ - plt = (Elf32_Addr *) (l->l_addr + l->l_info[DT_PLTGOT]->d_un.d_ptr); - - /* The beginning of the PLT does: - - save %sp, -64, %sp - pltpc: call _dl_runtime_resolve - nop - .word MAP - - This saves the register window containing the arguments, and the - PC value (pltpc) implicitly saved in %o7 by the call points near the - location where we store the link_map pointer for this object. */ - - plt[0] = OPCODE_SAVE_SP64; /* save %sp, -64, %sp */ - /* Construct PC-relative word address. */ - plt[1] = OPCODE_CALL | (((Elf32_Addr) &_dl_runtime_resolve - - (Elf32_Addr) &plt[1]) >> 2); - plt[2] = OPCODE_NOP; /* Fill call delay slot. */ - plt[3] = (Elf32_Addr *) l; - } - - return lazy; -} - -/* This code is used in dl-runtime.c to call the `fixup' function - and then redirect to the address it returns. */ -#define ELF_MACHINE_RUNTIME_TRAMPOLINE asm ("\ -# Trampoline for _dl_runtime_resolver - .globl _dl_runtime_resolve - .type _dl_runtime_resolve, @function -_dl_runtime_resolve: - t 1 - #call %g0 - # Pass two args to fixup: the PLT address computed from the PC saved - # in the PLT's call insn, and the reloc offset passed in %g1. - #ld [%o7 + 8], %o1 | Second arg, loaded from PLTPC[2]. - #call fixup - #shrl %g1, 22, %o0 | First arg, set in delay slot of call. - # Jump to the real function. - #jmpl %o0, %g0 - # In the delay slot of that jump, restore the register window - # saved by the first insn of the PLT. - #restore - .size _dl_runtime_resolve, . - _dl_runtime_resolve -"); - -/* The PLT uses Elf32_Rela relocs. */ -#define elf_machine_relplt elf_machine_rela - - -/* Mask identifying addresses reserved for the user program, - where the dynamic linker should not map anything. */ -#define ELF_MACHINE_USER_ADDRESS_MASK ??? - -/* Initial entry point code for the dynamic linker. - The C function `_dl_start' is the real entry point; - its return value is the user program's entry point. */ - -#define RTLD_START __asm__ ( \ -".text\n\ - .globl _start\n\ - .type _start,@function\n\ -_start:\n\ - /* Pass pointer to argument block to _dl_start. */\n\ - add %sp,64,%o0\n\ - call _dl_start\n\ - nop\n\ - \n\ - mov %o0,%l0\n\ - \n\ -2:\n\ - call 1f\n\ - nop\n\ -1:\n\ - sethi %hi(_GLOBAL_OFFSET_TABLE_-(2b-.)),%l2\n\ - sethi %hi(_dl_default_scope),%l3\n\ - or %l2,%lo(_GLOBAL_OFFSET_TABLE_-(2b-.)),%l2\n\ - or %l3,%lo(_dl_default_scope),%l3\n\ - add %o7,%l2,%l1\n\ - # %l1 has the GOT. %l3 has _dl_default_scope GOT offset\n\ - ld [%l1+%l3],%l4\n\ - # %l4 has pointer to _dl_default_scope. Now, load _dl_default_scope [2]\n\ - ld [%l4+8],%l4\n\ - # %l4 has _dl_default_scope [2]\n\ - # call _dl_init_next until it returns 0, pass _dl_default_scope [2]\n\ -3:\n\ - call _dl_init_next\n\ - mov %l4,%o0\n\ - cmp %o0,%g0\n\ - bz,a 4f\n\ - nop\n\ - call %o0\n\ - /* Pass pointer to argument block to this init function */\n\ - add %sp,64,%o0\n\ - b,a 3b\n\ -4:\n\ - # Clear the _dl_starting_up variable and pass _dl_fini in %g1 as per ELF ABI.\n\ - sethi %hi(_dl_starting_up),%l4\n\ - sethi %hi(_dl_fini),%l3\n\ - or %l4,%lo(_dl_starting_up),%l4\n\ - or %l3,%lo(_dl_fini),%l3\n\ - # clear _dl_starting_up\n\ - ld [%l1+%l4],%l5\n\ - st %g0,[%l5]\n\ - # load out fini function for atexit in %g1\n\ - ld [%l3+%l1],%g1\n\ - # jump to the user program entry point.\n\ - jmpl %l0,%g0\n\ - nop\n\ -"); diff --git a/sysdeps/sparc/dotmul.S b/sysdeps/sparc/dotmul.S deleted file mode 100644 index 7ce695cb7a..0000000000 --- a/sysdeps/sparc/dotmul.S +++ /dev/null @@ -1,123 +0,0 @@ -/* - * Signed multiply, from Appendix E of the Sparc Version 8 - * Architecture Manual. - */ - -/* - * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of - * the 64-bit product). - * - * This code optimizes short (less than 13-bit) multiplies. - */ - -#include "sysdep.h" -ENTRY(.mul) - mov %o0, %y ! multiplier -> Y - andncc %o0, 0xfff, %g0 ! test bits 12..31 - be Lmul_shortway ! if zero, can do it the short way - andcc %g0, %g0, %o4 ! zero the partial product and clear N and V - - /* - * Long multiply. 32 steps, followed by a final shift step. - */ - mulscc %o4, %o1, %o4 ! 1 - mulscc %o4, %o1, %o4 ! 2 - mulscc %o4, %o1, %o4 ! 3 - mulscc %o4, %o1, %o4 ! 4 - mulscc %o4, %o1, %o4 ! 5 - mulscc %o4, %o1, %o4 ! 6 - mulscc %o4, %o1, %o4 ! 7 - mulscc %o4, %o1, %o4 ! 8 - mulscc %o4, %o1, %o4 ! 9 - mulscc %o4, %o1, %o4 ! 10 - mulscc %o4, %o1, %o4 ! 11 - mulscc %o4, %o1, %o4 ! 12 - mulscc %o4, %o1, %o4 ! 13 - mulscc %o4, %o1, %o4 ! 14 - mulscc %o4, %o1, %o4 ! 15 - mulscc %o4, %o1, %o4 ! 16 - mulscc %o4, %o1, %o4 ! 17 - mulscc %o4, %o1, %o4 ! 18 - mulscc %o4, %o1, %o4 ! 19 - mulscc %o4, %o1, %o4 ! 20 - mulscc %o4, %o1, %o4 ! 21 - mulscc %o4, %o1, %o4 ! 22 - mulscc %o4, %o1, %o4 ! 23 - mulscc %o4, %o1, %o4 ! 24 - mulscc %o4, %o1, %o4 ! 25 - mulscc %o4, %o1, %o4 ! 26 - mulscc %o4, %o1, %o4 ! 27 - mulscc %o4, %o1, %o4 ! 28 - mulscc %o4, %o1, %o4 ! 29 - mulscc %o4, %o1, %o4 ! 30 - mulscc %o4, %o1, %o4 ! 31 - mulscc %o4, %o1, %o4 ! 32 - mulscc %o4, %g0, %o4 ! final shift - - ! If %o0 was negative, the result is - ! (%o0 * %o1) + (%o1 << 32)) - ! We fix that here. - -#if 0 - tst %o0 - bge 1f - rd %y, %o0 - - ! %o0 was indeed negative; fix upper 32 bits of result by subtracting - ! %o1 (i.e., return %o4 - %o1 in %o1). - retl - sub %o4, %o1, %o1 - -1: - retl - mov %o4, %o1 -#else - /* Faster code adapted from tege@sics.se's code for umul.S. */ - sra %o0, 31, %o2 ! make mask from sign bit - and %o1, %o2, %o2 ! %o2 = 0 or %o1, depending on sign of %o0 - rd %y, %o0 ! get lower half of product - retl - sub %o4, %o2, %o1 ! subtract compensation - ! and put upper half in place -#endif - -Lmul_shortway: - /* - * Short multiply. 12 steps, followed by a final shift step. - * The resulting bits are off by 12 and (32-12) = 20 bit positions, - * but there is no problem with %o0 being negative (unlike above). - */ - mulscc %o4, %o1, %o4 ! 1 - mulscc %o4, %o1, %o4 ! 2 - mulscc %o4, %o1, %o4 ! 3 - mulscc %o4, %o1, %o4 ! 4 - mulscc %o4, %o1, %o4 ! 5 - mulscc %o4, %o1, %o4 ! 6 - mulscc %o4, %o1, %o4 ! 7 - mulscc %o4, %o1, %o4 ! 8 - mulscc %o4, %o1, %o4 ! 9 - mulscc %o4, %o1, %o4 ! 10 - mulscc %o4, %o1, %o4 ! 11 - mulscc %o4, %o1, %o4 ! 12 - mulscc %o4, %g0, %o4 ! final shift - - /* - * %o4 has 20 of the bits that should be in the low part of the - * result; %y has the bottom 12 (as %y's top 12). That is: - * - * %o4 %y - * +----------------+----------------+ - * | -12- | -20- | -12- | -20- | - * +------(---------+------)---------+ - * --hi-- ----low-part---- - * - * The upper 12 bits of %o4 should be sign-extended to form the - * high part of the product (i.e., highpart = %o4 >> 20). - */ - - rd %y, %o5 - sll %o4, 12, %o0 ! shift middle bits left 12 - srl %o5, 20, %o5 ! shift low bits right 20, zero fill at left - or %o5, %o0, %o0 ! construct low part of result - retl - sra %o4, 20, %o1 ! ... and extract high part of result diff --git a/sysdeps/sparc/e_sqrt.c b/sysdeps/sparc/e_sqrt.c deleted file mode 100644 index a98ae21ace..0000000000 --- a/sysdeps/sparc/e_sqrt.c +++ /dev/null @@ -1,34 +0,0 @@ -/* Copyright (C) 1991, 1992, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include -#include - -#ifndef __GNUC__ - #error This file uses GNU C extensions; you must compile with GCC. -#endif - -/* Return the square root of X. */ -double -__ieee754_sqrt (x) - double x; -{ - register double result; - asm ("fsqrtd %1, %0" : "=f" (result) : "f" (x)); - return result; -} diff --git a/sysdeps/sparc/elf/Makefile b/sysdeps/sparc/elf/Makefile deleted file mode 100644 index 319fbdef10..0000000000 --- a/sysdeps/sparc/elf/Makefile +++ /dev/null @@ -1,4 +0,0 @@ -# Sparc/ELF specific definitions. - -# The assembler on SPARC needs the -fPIC flag even when it's assembler code. -ASFLAGS-.so = -fPIC diff --git a/sysdeps/sparc/elf/start.c b/sysdeps/sparc/elf/start.c deleted file mode 100644 index f1e80195ca..0000000000 --- a/sysdeps/sparc/elf/start.c +++ /dev/null @@ -1,68 +0,0 @@ -/* Copyright (C) 1991, 1992, 1993, 1994, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include -#include -#include -#include - -extern char **__environ; - -extern void __libc_init_first __P ((int argc, char **argv, char **envp)); -extern int main __P ((int argc, char **argv, char **envp)); - -register long int sp asm("%sp"), fp asm("%fp"); - -void -_start (void) -{ - /* It is important that these be declared `register'. - Otherwise, when compiled without optimization, they are put on the - stack, which loses completely after we zero the FP. */ - register int argc; - register char **argv, **envp; - register long int g1 asm ("%g1"); - unsigned long int copy_g1 = g1; - - /* Unwind the frame built when we entered the function. */ - asm("restore"); - if (copy_g1) - atexit (copy_g1); - - /* And clear the frame pointer. */ - fp = 0; - - /* The argument info starts after one register - window (64 bytes) past the SP. */ - argc = ((int *) sp)[16]; - argv = (char **) &((int *) sp)[17]; - envp = &argv[argc + 1]; - __environ = envp; - - /* Allocate 24 bytes of stack space for the register save area. */ - sp -= 24; - __libc_init_first (argc, argv, envp); -#ifdef ELF_INIT_FINI - { - extern void _fini (void); - _init (); - atexit (_fini); - } -#endif - exit (main (argc, argv, envp)); -} diff --git a/sysdeps/sparc/fpu/bits/mathdef.h b/sysdeps/sparc/fpu/bits/mathdef.h new file mode 100644 index 0000000000..0bc9c94ecc --- /dev/null +++ b/sysdeps/sparc/fpu/bits/mathdef.h @@ -0,0 +1,64 @@ +/* Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _MATH_H +#error "Never use directly; include instead" +#endif + +/* FIXME! This file describes properties of the compiler, not the machine; + it should not be part of libc! */ + +#ifdef __GNUC__ +#if __STDC__ == 1 + +/* In GNU or ANSI mode, gcc leaves `float' expressions as-is. */ +typedef float float_t; +typedef double double_t; + +/* Signal that types stay as they were declared. */ +#define FLT_EVAL_METHOD 0 + +/* Define `INFINITY' as value of type `float_t'. */ +#define INFINITY HUGE_VALF + +#else + +/* For `gcc -traditional', `float' expressions are evaluated as `double'. */ +typedef double float_t; +typedef double double_t; + +/* Signal that both types are `double'. */ +#define FLT_EVAL_METHOD 1 + +/* Define `INFINITY' as value of type `float_t'. */ +#define INFINITY HUGE_VAL + +#endif +#else + +/* Wild guess at types for float_t and double_t. */ +typedef double float_t; +typedef double double_t; + +/* Strange compiler, we don't know how it works. */ +#define FLT_EVAL_METHOD -1 + +/* Define `INFINITY' as value of type `float_t'. */ +#define INFINITY HUGE_VAL + +#endif diff --git a/sysdeps/sparc/fpu/fclrexcpt.c b/sysdeps/sparc/fpu/fclrexcpt.c new file mode 100644 index 0000000000..8dc68e84e9 --- /dev/null +++ b/sysdeps/sparc/fpu/fclrexcpt.c @@ -0,0 +1,32 @@ +/* Clear given exceptions in current floating-point environment. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +void +feclearexcept (int excepts) +{ + fenv_t tmp; + + __fenv_stfsr (tmp); + + tmp &= ~(excepts & FE_ALL_EXCEPT); + + __fenv_ldfsr (tmp); +} diff --git a/sysdeps/sparc/fpu/fegetenv.c b/sysdeps/sparc/fpu/fegetenv.c new file mode 100644 index 0000000000..a796ccb731 --- /dev/null +++ b/sysdeps/sparc/fpu/fegetenv.c @@ -0,0 +1,26 @@ +/* Store current floating-point environment. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +void +fegetenv (fenv_t *envp) +{ + __fenv_stfsr (*envp); +} diff --git a/sysdeps/sparc/fpu/fegetround.c b/sysdeps/sparc/fpu/fegetround.c new file mode 100644 index 0000000000..ea1df2b6d8 --- /dev/null +++ b/sysdeps/sparc/fpu/fegetround.c @@ -0,0 +1,30 @@ +/* Return current rounding direction. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +int +fegetround (void) +{ + fenv_t tmp; + + __fenv_stfsr (tmp); + + return tmp & __FE_ROUND_MASK; +} diff --git a/sysdeps/sparc/fpu/feholdexcpt.c b/sysdeps/sparc/fpu/feholdexcpt.c new file mode 100644 index 0000000000..9b4c2134f7 --- /dev/null +++ b/sysdeps/sparc/fpu/feholdexcpt.c @@ -0,0 +1,35 @@ +/* Store current floating-point environment and clear exceptions. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +int +feholdexcept (fenv_t *envp) +{ + fenv_t tmp; + + __fenv_stfsr (*envp); + + /* Set all exceptions to non-stop. */ + tmp = *envp | (0x1f << 23); + + __fenv_ldfsr (tmp); + + return 1; +} diff --git a/sysdeps/sparc/fpu/fesetenv.c b/sysdeps/sparc/fpu/fesetenv.c new file mode 100644 index 0000000000..44c5b2629c --- /dev/null +++ b/sysdeps/sparc/fpu/fesetenv.c @@ -0,0 +1,43 @@ +/* Install given floating-point environment. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +void +fesetenv (const fenv_t *envp) +{ + fenv_t dummy; + + /* Put these constants in memory explicitly, so as to cope with a + -fPIC bug as of gcc 970624. Making them automatic is quicker + than loading up the pic register in this instance. */ + + if (envp == FE_DFL_ENV) + { + dummy = 0; + envp = &dummy; + } + else if (envp == FE_NOMASK_ENV) + { + dummy = 0x1f << 23; + envp = &dummy; + } + + __fenv_ldfsr (*envp); +} diff --git a/sysdeps/sparc/fpu/fesetround.c b/sysdeps/sparc/fpu/fesetround.c new file mode 100644 index 0000000000..74963d7573 --- /dev/null +++ b/sysdeps/sparc/fpu/fesetround.c @@ -0,0 +1,37 @@ +/* Set current rounding direction. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +int +fesetround (int round) +{ + fenv_t tmp; + + if ((round & ~__FE_ROUND_MASK) != 0) + /* ROUND is no valid rounding mode. */ + return 0; + + __fenv_stfsr (tmp); + tmp &= ~__FE_ROUND_MASK; + tmp |= round; + __fenv_ldfsr (tmp); + + return 1; +} diff --git a/sysdeps/sparc/fpu/feupdateenv.c b/sysdeps/sparc/fpu/feupdateenv.c new file mode 100644 index 0000000000..0fba9975f8 --- /dev/null +++ b/sysdeps/sparc/fpu/feupdateenv.c @@ -0,0 +1,38 @@ +/* Install given floating-point environment and raise exceptions. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +void +feupdateenv (const fenv_t *envp) +{ + fexcept_t tmp; + + /* Save current exceptions. */ + __fenv_stfsr (tmp); + tmp &= FE_ALL_EXCEPT; + + /* Install new environment. */ + fesetenv (envp); + + /* Raise the safed exception. Incidently for us the implementation + defined format of the values in objects of type fexcept_t is the + same as the ones specified using the FE_* constants. */ + feraiseexcept ((int) tmp); +} diff --git a/sysdeps/sparc/fpu/fgetexcptflg.c b/sysdeps/sparc/fpu/fgetexcptflg.c new file mode 100644 index 0000000000..977c93e2ca --- /dev/null +++ b/sysdeps/sparc/fpu/fgetexcptflg.c @@ -0,0 +1,31 @@ +/* Store current representation for exceptions. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +void +fegetexceptflag (fexcept_t *flagp, int excepts) +{ + fexcept_t tmp; + + /* Get the current exceptions. */ + __fenv_stfsr (tmp); + + *flagp = tmp & excepts & FE_ALL_EXCEPT; +} diff --git a/sysdeps/sparc/fpu/fraiseexcpt.c b/sysdeps/sparc/fpu/fraiseexcpt.c new file mode 100644 index 0000000000..308c9b2e7f --- /dev/null +++ b/sysdeps/sparc/fpu/fraiseexcpt.c @@ -0,0 +1,66 @@ +/* Raise given exceptions. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include +#include + +static void +ignore_me(double foo) +{ +} + +void +feraiseexcept (int excepts) +{ + /* Raise exceptions represented by EXPECTS. But we must raise only + one signal at a time. It is important the if the overflow/underflow + exception and the inexact exception are given at the same time, + the overflow/underflow exception follows the inexact exception. */ + + /* First: invalid exception. */ + if ((FE_INVALID & excepts) != 0) + { + /* One example of a invalid operation is 0/0. */ + ignore_me (0.0 / 0.0); + } + + /* Next: division by zero. */ + if ((FE_DIVBYZERO & excepts) != 0) + { + ignore_me (1.0 / 0.0); + } + + /* Next: overflow. */ + if ((FE_OVERFLOW & excepts) != 0) + { + ignore_me (LDBL_MAX * LDBL_MAX); + } + + /* Next: underflow. */ + if ((FE_UNDERFLOW & excepts) != 0) + { + ignore_me (LDBL_MIN / 16.0); + } + + /* Last: inexact. */ + if ((FE_INEXACT & excepts) != 0) + { + ignore_me (1.0 / M_PI); + } +} diff --git a/sysdeps/sparc/fpu/fsetexcptflg.c b/sysdeps/sparc/fpu/fsetexcptflg.c new file mode 100644 index 0000000000..1029cde03d --- /dev/null +++ b/sysdeps/sparc/fpu/fsetexcptflg.c @@ -0,0 +1,34 @@ +/* Set floating-point environment exception handling. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include +#include + +void +fesetexceptflag (const fexcept_t *flagp, int excepts) +{ + fenv_t tmp; + + __fenv_stfsr (tmp); + + tmp &= ~(excepts & FE_ALL_EXCEPT); + tmp |= *flagp & excepts & FE_ALL_EXCEPT; + + __fenv_ldfsr (tmp); +} diff --git a/sysdeps/sparc/fpu/ftestexcept.c b/sysdeps/sparc/fpu/ftestexcept.c new file mode 100644 index 0000000000..3efe9e896b --- /dev/null +++ b/sysdeps/sparc/fpu/ftestexcept.c @@ -0,0 +1,30 @@ +/* Test exception in current environment. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +int +fetestexcept (int excepts) +{ + int tmp; + + __fenv_stfsr (tmp); + + return tmp & excepts & FE_ALL_EXCEPT; +} diff --git a/sysdeps/sparc/fpu_control.h b/sysdeps/sparc/fpu_control.h deleted file mode 100644 index 5ef3824702..0000000000 --- a/sysdeps/sparc/fpu_control.h +++ /dev/null @@ -1,69 +0,0 @@ -/* FPU control word bits. SPARC version. - Copyright (C) 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Miguel de Icaza - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#ifndef _FPU_CONTROL_H -#define _FPU_CONTROL_H 1 - - -#include - -/* precision control */ -#define _FPU_EXTENDED 0x00000000 /* RECOMMENDED */ -#define _FPU_DOUBLE 0x20000000 -#define _FPU_80BIT 0x30000000 -#define _FPU_SINGLE 0x10000000 /* DO NOT USE */ - -/* rounding control / Sparc */ -#define _FPU_RC_DOWN 0xc0000000 -#define _FPU_RC_UP 0x80000000 -#define _FPU_RC_ZERO 0x40000000 -#define _FPU_RC_NEAREST 0x0 /* RECOMMENDED */ - -#define _FPU_RESERVED 0x30300000 /* Reserved bits in cw */ - - -/* Now two recommended cw */ - -/* Linux default: - - extended precision - - rounding to nearest - - exceptions on overflow, zero divide and NaN */ -#define _FPU_DEFAULT 0x1e - -/* IEEE: same as above, but exceptions */ -#define _FPU_IEEE 0x0 - -/* Type of the control word. */ -typedef unsigned int fpu_control_t; - -#define _FPU_GETCW(cw) __asm__ ("st %%fsr,%0" : "=m" (*&cw)) -#define _FPU_SETCW(cw) __asm__ ("ld %0,%%fsr" : : "m" (*&cw)) - -/* Default control word set at startup. */ -extern fpu_control_t __fpu_control; - -__BEGIN_DECLS - -/* Called at startup. It can be used to manipulate fpu control register. */ -extern void __setfpucw __P ((fpu_control_t)); - -__END_DECLS - -#endif /* fpu_control.h */ diff --git a/sysdeps/sparc/lshift.S b/sysdeps/sparc/lshift.S deleted file mode 100644 index 4f0595f2fb..0000000000 --- a/sysdeps/sparc/lshift.S +++ /dev/null @@ -1,95 +0,0 @@ -! sparc __mpn_lshift -- - -! Copyright (C) 1995, 1996 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr %o0 -! src_ptr %o1 -! size %o2 -! cnt %o3 - -#include "sysdep.h" - - .text - .align 4 - .global C_SYMBOL_NAME(__mpn_lshift) -C_SYMBOL_NAME(__mpn_lshift): - sll %o2,2,%g1 - add %o1,%g1,%o1 ! make %o1 point at end of src - ld [%o1-4],%g2 ! load first limb - sub %g0,%o3,%o5 ! negate shift count - add %o0,%g1,%o0 ! make %o0 point at end of res - add %o2,-1,%o2 - andcc %o2,4-1,%g4 ! number of limbs in first loop - srl %g2,%o5,%g1 ! compute function result - be L0 ! if multiple of 4 limbs, skip first loop - st %g1,[%sp+80] - - sub %o2,%g4,%o2 ! adjust count for main loop - -Loop0: ld [%o1-8],%g3 - add %o0,-4,%o0 - add %o1,-4,%o1 - addcc %g4,-1,%g4 - sll %g2,%o3,%o4 - srl %g3,%o5,%g1 - mov %g3,%g2 - or %o4,%g1,%o4 - bne Loop0 - st %o4,[%o0+0] - -L0: tst %o2 - be Lend - nop - -Loop: ld [%o1-8],%g3 - add %o0,-16,%o0 - addcc %o2,-4,%o2 - sll %g2,%o3,%o4 - srl %g3,%o5,%g1 - - ld [%o1-12],%g2 - sll %g3,%o3,%g4 - or %o4,%g1,%o4 - st %o4,[%o0+12] - srl %g2,%o5,%g1 - - ld [%o1-16],%g3 - sll %g2,%o3,%o4 - or %g4,%g1,%g4 - st %g4,[%o0+8] - srl %g3,%o5,%g1 - - ld [%o1-20],%g2 - sll %g3,%o3,%g4 - or %o4,%g1,%o4 - st %o4,[%o0+4] - srl %g2,%o5,%g1 - - add %o1,-16,%o1 - or %g4,%g1,%g4 - bne Loop - st %g4,[%o0+0] - -Lend: sll %g2,%o3,%g2 - st %g2,[%o0-4] - retl - ld [%sp+80],%o0 diff --git a/sysdeps/sparc/memcopy.h b/sysdeps/sparc/memcopy.h deleted file mode 100644 index ff73f453da..0000000000 --- a/sysdeps/sparc/memcopy.h +++ /dev/null @@ -1,21 +0,0 @@ -/* Copyright (C) 1991, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include -#undef reg_char -#define reg_char int diff --git a/sysdeps/sparc/mul_1.S b/sysdeps/sparc/mul_1.S deleted file mode 100644 index 142fd8ba2a..0000000000 --- a/sysdeps/sparc/mul_1.S +++ /dev/null @@ -1,199 +0,0 @@ -! SPARC __mpn_mul_1 -- Multiply a limb vector with a limb and store -! the result in a second limb vector. - -! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr o0 -! s1_ptr o1 -! size o2 -! s2_limb o3 - -! ADD CODE FOR SMALL MULTIPLIERS! -!1: ld -! st -! -!2: ld ,a -! addxcc a,a,x -! st x, -! -!3_unrolled: -! ld ,a -! addxcc a,a,x1 ! 2a + cy -! addx %g0,%g0,x2 -! addcc a,x1,x ! 3a + c -! st x, -! -! ld ,a -! addxcc a,a,y1 -! addx %g0,%g0,y2 -! addcc a,y1,x -! st x, -! -!4_unrolled: -! ld ,a -! srl a,2,x1 ! 4a -! addxcc y2,x1,x -! sll a,30,x2 -! st x, -! -! ld ,a -! srl a,2,y1 -! addxcc x2,y1,y -! sll a,30,y2 -! st x, -! -!5_unrolled: -! ld ,a -! srl a,2,x1 ! 4a -! addxcc a,x1,x ! 5a + c -! sll a,30,x2 -! addx %g0,x2,x2 -! st x, -! -! ld ,a -! srl a,2,y1 -! addxcc a,y1,x -! sll a,30,y2 -! addx %g0,y2,y2 -! st x, -! -!8_unrolled: -! ld ,a -! srl a,3,x1 ! 8a -! addxcc y2,x1,x -! sll a,29,x2 -! st x, -! -! ld ,a -! srl a,3,y1 -! addxcc x2,y1,y -! sll a,29,y2 -! st x, - -#include "sysdep.h" - -.text - .align 4 - .global C_SYMBOL_NAME(__mpn_mul_1) -C_SYMBOL_NAME(__mpn_mul_1): - ! Make S1_PTR and RES_PTR point at the end of their blocks - ! and put (- 4 x SIZE) in index/loop counter. - sll %o2,2,%o2 - add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval - add %o1,%o2,%o1 - sub %g0,%o2,%o2 - - cmp %o3,0xfff - bgu Large - nop - - ld [%o1+%o2],%o5 - mov 0,%o0 - b L0 - add %o4,-4,%o4 -Loop0: - st %g1,[%o4+%o2] -L0: wr %g0,%o3,%y - sra %o5,31,%g2 - and %o3,%g2,%g2 - andcc %g1,0,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,0,%g1 - sra %g1,20,%g4 - sll %g1,12,%g1 - rd %y,%g3 - srl %g3,20,%g3 - or %g1,%g3,%g1 - - addcc %g1,%o0,%g1 - addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb - addcc %o2,4,%o2 ! loop counter - bne,a Loop0 - ld [%o1+%o2],%o5 - - retl - st %g1,[%o4+%o2] - - -Large: ld [%o1+%o2],%o5 - mov 0,%o0 - sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 - b L1 - add %o4,-4,%o4 -Loop: - st %g3,[%o4+%o2] -L1: wr %g0,%o5,%y - and %o5,%g4,%g2 ! g2 = S1_LIMB iff S2_LIMB < 0, else 0 - andcc %g0,%g0,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%g0,%g1 - rd %y,%g3 - addcc %g3,%o0,%g3 - addx %g2,%g1,%o0 ! add sign-compensation and cy to hi limb - addcc %o2,4,%o2 ! loop counter - bne,a Loop - ld [%o1+%o2],%o5 - - retl - st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/rem.S b/sysdeps/sparc/rem.S deleted file mode 100644 index d50f1af630..0000000000 --- a/sysdeps/sparc/rem.S +++ /dev/null @@ -1,369 +0,0 @@ - /* This file is generated from divrem.m4; DO NOT EDIT! */ -/* - * Division and remainder, from Appendix E of the Sparc Version 8 - * Architecture Manual, with fixes from Gordon Irlam. - */ - -/* - * Input: dividend and divisor in %o0 and %o1 respectively. - * - * m4 parameters: - * .rem name of function to generate - * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 - * true true=true => signed; true=false => unsigned - * - * Algorithm parameters: - * N how many bits per iteration we try to get (4) - * WORDSIZE total number of bits (32) - * - * Derived constants: - * TOPBITS number of bits in the top decade of a number - * - * Important variables: - * Q the partial quotient under development (initially 0) - * R the remainder so far, initially the dividend - * ITER number of main division loop iterations required; - * equal to ceil(log2(quotient) / N). Note that this - * is the log base (2^N) of the quotient. - * V the current comparand, initially divisor*2^(ITER*N-1) - * - * Cost: - * Current estimate for non-large dividend is - * ceil(log2(quotient) / N) * (10 + 7N/2) + C - * A large dividend is one greater than 2^(31-TOPBITS) and takes a - * different path, as the upper bits of the quotient must be developed - * one bit at a time. - */ - - - -#include "sysdep.h" -#ifdef __linux__ -#include -#else -#ifdef __svr4__ -#include -#else -#include -#endif -#endif - -ENTRY(.rem) - ! compute sign of result; if neither is negative, no problem - orcc %o1, %o0, %g0 ! either negative? - bge 2f ! no, go do the divide - mov %o0, %g6 ! sign of remainder matches %o0 - tst %o1 - bge 1f - tst %o0 - ! %o1 is definitely negative; %o0 might also be negative - bge 2f ! if %o0 not negative... - sub %g0, %o1, %o1 ! in any case, make %o1 nonneg -1: ! %o0 is negative, %o1 is nonnegative - sub %g0, %o0, %o0 ! make %o0 nonnegative -2: - - ! Ready to divide. Compute size of quotient; scale comparand. - orcc %o1, %g0, %o5 - bne 1f - mov %o0, %o3 - - ! Divide by zero trap. If it returns, return 0 (about as - ! wrong as possible, but that is what SunOS does...). - ta ST_DIV0 - retl - clr %o0 - -1: - cmp %o3, %o5 ! if %o1 exceeds %o0, done - blu Lgot_result ! (and algorithm fails otherwise) - clr %o2 - sethi %hi(1 << (32 - 4 - 1)), %g1 - cmp %o3, %g1 - blu Lnot_really_big - clr %o4 - - ! Here the dividend is >= 2**(31-N) or so. We must be careful here, - ! as our usual N-at-a-shot divide step will cause overflow and havoc. - ! The number of bits in the result here is N*ITER+SC, where SC <= N. - ! Compute ITER in an unorthodox manner: know we need to shift V into - ! the top decade: so do not even bother to compare to R. - 1: - cmp %o5, %g1 - bgeu 3f - mov 1, %g7 - sll %o5, 4, %o5 - b 1b - add %o4, 1, %o4 - - ! Now compute %g7. - 2: addcc %o5, %o5, %o5 - bcc Lnot_too_big - add %g7, 1, %g7 - - ! We get here if the %o1 overflowed while shifting. - ! This means that %o3 has the high-order bit set. - ! Restore %o5 and subtract from %o3. - sll %g1, 4, %g1 ! high order bit - srl %o5, 1, %o5 ! rest of %o5 - add %o5, %g1, %o5 - b Ldo_single_div - sub %g7, 1, %g7 - - Lnot_too_big: - 3: cmp %o5, %o3 - blu 2b - nop - be Ldo_single_div - nop - /* NB: these are commented out in the V8-Sparc manual as well */ - /* (I do not understand this) */ - ! %o5 > %o3: went too far: back up 1 step - ! srl %o5, 1, %o5 - ! dec %g7 - ! do single-bit divide steps - ! - ! We have to be careful here. We know that %o3 >= %o5, so we can do the - ! first divide step without thinking. BUT, the others are conditional, - ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- - ! order bit set in the first step, just falling into the regular - ! division loop will mess up the first time around. - ! So we unroll slightly... - Ldo_single_div: - subcc %g7, 1, %g7 - bl Lend_regular_divide - nop - sub %o3, %o5, %o3 - mov 1, %o2 - b Lend_single_divloop - nop - Lsingle_divloop: - sll %o2, 1, %o2 - bl 1f - srl %o5, 1, %o5 - ! %o3 >= 0 - sub %o3, %o5, %o3 - b 2f - add %o2, 1, %o2 - 1: ! %o3 < 0 - add %o3, %o5, %o3 - sub %o2, 1, %o2 - 2: - Lend_single_divloop: - subcc %g7, 1, %g7 - bge Lsingle_divloop - tst %o3 - b,a Lend_regular_divide - -Lnot_really_big: -1: - sll %o5, 4, %o5 - cmp %o5, %o3 - bleu 1b - addcc %o4, 1, %o4 - be Lgot_result - sub %o4, 1, %o4 - - tst %o3 ! set up for initial iteration -Ldivloop: - sll %o2, 4, %o2 - ! depth 1, accumulated bits 0 - bl L.1.16 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 2, accumulated bits 1 - bl L.2.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits 3 - bl L.3.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 7 - bl L.4.23 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (7*2+1), %o2 - -L.4.23: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (7*2-1), %o2 - - -L.3.19: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 5 - bl L.4.21 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (5*2+1), %o2 - -L.4.21: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (5*2-1), %o2 - - - -L.2.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits 1 - bl L.3.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 3 - bl L.4.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (3*2+1), %o2 - -L.4.19: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (3*2-1), %o2 - - -L.3.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 1 - bl L.4.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (1*2+1), %o2 - -L.4.17: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (1*2-1), %o2 - - - - -L.1.16: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 2, accumulated bits -1 - bl L.2.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits -1 - bl L.3.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -1 - bl L.4.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2+1), %o2 - -L.4.15: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2-1), %o2 - - -L.3.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -3 - bl L.4.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2+1), %o2 - -L.4.13: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2-1), %o2 - - - -L.2.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits -3 - bl L.3.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -5 - bl L.4.11 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2+1), %o2 - -L.4.11: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2-1), %o2 - - -L.3.13: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -7 - bl L.4.9 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2+1), %o2 - -L.4.9: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2-1), %o2 - - - - - 9: -Lend_regular_divide: - subcc %o4, 1, %o4 - bge Ldivloop - tst %o3 - bl,a Lgot_result - ! non-restoring fixup here (one instruction only!) - add %o3, %o1, %o3 - - -Lgot_result: - ! check to see if answer should be < 0 - tst %g6 - bl,a 1f - sub %g0, %o3, %o3 -1: - retl - mov %o3, %o0 diff --git a/sysdeps/sparc/rshift.S b/sysdeps/sparc/rshift.S deleted file mode 100644 index fea4f3b926..0000000000 --- a/sysdeps/sparc/rshift.S +++ /dev/null @@ -1,92 +0,0 @@ -! sparc __mpn_rshift -- - -! Copyright (C) 1995, 1996 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr %o0 -! src_ptr %o1 -! size %o2 -! cnt %o3 - -#include "sysdep.h" - - .text - .align 4 - .global C_SYMBOL_NAME(__mpn_rshift) -C_SYMBOL_NAME(__mpn_rshift): - ld [%o1],%g2 ! load first limb - sub %g0,%o3,%o5 ! negate shift count - add %o2,-1,%o2 - andcc %o2,4-1,%g4 ! number of limbs in first loop - sll %g2,%o5,%g1 ! compute function result - be L0 ! if multiple of 4 limbs, skip first loop - st %g1,[%sp+80] - - sub %o2,%g4,%o2 ! adjust count for main loop - -Loop0: ld [%o1+4],%g3 - add %o0,4,%o0 - add %o1,4,%o1 - addcc %g4,-1,%g4 - srl %g2,%o3,%o4 - sll %g3,%o5,%g1 - mov %g3,%g2 - or %o4,%g1,%o4 - bne Loop0 - st %o4,[%o0-4] - -L0: tst %o2 - be Lend - nop - -Loop: ld [%o1+4],%g3 - add %o0,16,%o0 - addcc %o2,-4,%o2 - srl %g2,%o3,%o4 - sll %g3,%o5,%g1 - - ld [%o1+8],%g2 - srl %g3,%o3,%g4 - or %o4,%g1,%o4 - st %o4,[%o0-16] - sll %g2,%o5,%g1 - - ld [%o1+12],%g3 - srl %g2,%o3,%o4 - or %g4,%g1,%g4 - st %g4,[%o0-12] - sll %g3,%o5,%g1 - - ld [%o1+16],%g2 - srl %g3,%o3,%g4 - or %o4,%g1,%o4 - st %o4,[%o0-8] - sll %g2,%o5,%g1 - - add %o1,16,%o1 - or %g4,%g1,%g4 - bne Loop - st %g4,[%o0-4] - -Lend: srl %g2,%o3,%g2 - st %g2,[%o0-0] - retl - ld [%sp+80],%o0 diff --git a/sysdeps/sparc/sdiv.S b/sysdeps/sparc/sdiv.S deleted file mode 100644 index 02ed2e973c..0000000000 --- a/sysdeps/sparc/sdiv.S +++ /dev/null @@ -1,369 +0,0 @@ - /* This file is generated from divrem.m4; DO NOT EDIT! */ -/* - * Division and remainder, from Appendix E of the Sparc Version 8 - * Architecture Manual, with fixes from Gordon Irlam. - */ - -/* - * Input: dividend and divisor in %o0 and %o1 respectively. - * - * m4 parameters: - * .div name of function to generate - * div div=div => %o0 / %o1; div=rem => %o0 % %o1 - * true true=true => signed; true=false => unsigned - * - * Algorithm parameters: - * N how many bits per iteration we try to get (4) - * WORDSIZE total number of bits (32) - * - * Derived constants: - * TOPBITS number of bits in the top decade of a number - * - * Important variables: - * Q the partial quotient under development (initially 0) - * R the remainder so far, initially the dividend - * ITER number of main division loop iterations required; - * equal to ceil(log2(quotient) / N). Note that this - * is the log base (2^N) of the quotient. - * V the current comparand, initially divisor*2^(ITER*N-1) - * - * Cost: - * Current estimate for non-large dividend is - * ceil(log2(quotient) / N) * (10 + 7N/2) + C - * A large dividend is one greater than 2^(31-TOPBITS) and takes a - * different path, as the upper bits of the quotient must be developed - * one bit at a time. - */ - - - -#include "sysdep.h" -#ifdef __linux__ -#include -#else -#ifdef __svr4__ -#include -#else -#include -#endif -#endif - -ENTRY(.div) - ! compute sign of result; if neither is negative, no problem - orcc %o1, %o0, %g0 ! either negative? - bge 2f ! no, go do the divide - xor %o1, %o0, %g6 ! compute sign in any case - tst %o1 - bge 1f - tst %o0 - ! %o1 is definitely negative; %o0 might also be negative - bge 2f ! if %o0 not negative... - sub %g0, %o1, %o1 ! in any case, make %o1 nonneg -1: ! %o0 is negative, %o1 is nonnegative - sub %g0, %o0, %o0 ! make %o0 nonnegative -2: - - ! Ready to divide. Compute size of quotient; scale comparand. - orcc %o1, %g0, %o5 - bne 1f - mov %o0, %o3 - - ! Divide by zero trap. If it returns, return 0 (about as - ! wrong as possible, but that is what SunOS does...). - ta ST_DIV0 - retl - clr %o0 - -1: - cmp %o3, %o5 ! if %o1 exceeds %o0, done - blu Lgot_result ! (and algorithm fails otherwise) - clr %o2 - sethi %hi(1 << (32 - 4 - 1)), %g1 - cmp %o3, %g1 - blu Lnot_really_big - clr %o4 - - ! Here the dividend is >= 2**(31-N) or so. We must be careful here, - ! as our usual N-at-a-shot divide step will cause overflow and havoc. - ! The number of bits in the result here is N*ITER+SC, where SC <= N. - ! Compute ITER in an unorthodox manner: know we need to shift V into - ! the top decade: so do not even bother to compare to R. - 1: - cmp %o5, %g1 - bgeu 3f - mov 1, %g7 - sll %o5, 4, %o5 - b 1b - add %o4, 1, %o4 - - ! Now compute %g7. - 2: addcc %o5, %o5, %o5 - bcc Lnot_too_big - add %g7, 1, %g7 - - ! We get here if the %o1 overflowed while shifting. - ! This means that %o3 has the high-order bit set. - ! Restore %o5 and subtract from %o3. - sll %g1, 4, %g1 ! high order bit - srl %o5, 1, %o5 ! rest of %o5 - add %o5, %g1, %o5 - b Ldo_single_div - sub %g7, 1, %g7 - - Lnot_too_big: - 3: cmp %o5, %o3 - blu 2b - nop - be Ldo_single_div - nop - /* NB: these are commented out in the V8-Sparc manual as well */ - /* (I do not understand this) */ - ! %o5 > %o3: went too far: back up 1 step - ! srl %o5, 1, %o5 - ! dec %g7 - ! do single-bit divide steps - ! - ! We have to be careful here. We know that %o3 >= %o5, so we can do the - ! first divide step without thinking. BUT, the others are conditional, - ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- - ! order bit set in the first step, just falling into the regular - ! division loop will mess up the first time around. - ! So we unroll slightly... - Ldo_single_div: - subcc %g7, 1, %g7 - bl Lend_regular_divide - nop - sub %o3, %o5, %o3 - mov 1, %o2 - b Lend_single_divloop - nop - Lsingle_divloop: - sll %o2, 1, %o2 - bl 1f - srl %o5, 1, %o5 - ! %o3 >= 0 - sub %o3, %o5, %o3 - b 2f - add %o2, 1, %o2 - 1: ! %o3 < 0 - add %o3, %o5, %o3 - sub %o2, 1, %o2 - 2: - Lend_single_divloop: - subcc %g7, 1, %g7 - bge Lsingle_divloop - tst %o3 - b,a Lend_regular_divide - -Lnot_really_big: -1: - sll %o5, 4, %o5 - cmp %o5, %o3 - bleu 1b - addcc %o4, 1, %o4 - be Lgot_result - sub %o4, 1, %o4 - - tst %o3 ! set up for initial iteration -Ldivloop: - sll %o2, 4, %o2 - ! depth 1, accumulated bits 0 - bl L.1.16 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 2, accumulated bits 1 - bl L.2.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits 3 - bl L.3.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 7 - bl L.4.23 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (7*2+1), %o2 - -L.4.23: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (7*2-1), %o2 - - -L.3.19: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 5 - bl L.4.21 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (5*2+1), %o2 - -L.4.21: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (5*2-1), %o2 - - - -L.2.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits 1 - bl L.3.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 3 - bl L.4.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (3*2+1), %o2 - -L.4.19: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (3*2-1), %o2 - - -L.3.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 1 - bl L.4.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (1*2+1), %o2 - -L.4.17: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (1*2-1), %o2 - - - - -L.1.16: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 2, accumulated bits -1 - bl L.2.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits -1 - bl L.3.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -1 - bl L.4.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2+1), %o2 - -L.4.15: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2-1), %o2 - - -L.3.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -3 - bl L.4.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2+1), %o2 - -L.4.13: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2-1), %o2 - - - -L.2.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits -3 - bl L.3.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -5 - bl L.4.11 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2+1), %o2 - -L.4.11: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2-1), %o2 - - -L.3.13: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -7 - bl L.4.9 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2+1), %o2 - -L.4.9: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2-1), %o2 - - - - - 9: -Lend_regular_divide: - subcc %o4, 1, %o4 - bge Ldivloop - tst %o3 - bl,a Lgot_result - ! non-restoring fixup here (one instruction only!) - sub %o2, 1, %o2 - - -Lgot_result: - ! check to see if answer should be < 0 - tst %g6 - bl,a 1f - sub %g0, %o2, %o2 -1: - retl - mov %o2, %o0 diff --git a/sysdeps/sparc/setjmp.S b/sysdeps/sparc/setjmp.S deleted file mode 100644 index 772ec723a3..0000000000 --- a/sysdeps/sparc/setjmp.S +++ /dev/null @@ -1,54 +0,0 @@ -/* Copyright (C) 1991, 1993, 1994, 1996, 1997 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -#include -#include - -#define _ASM 1 -#include - -ENTRY (__sigsetjmp) - /* Save our SP and FP; in the delay slot of the jump, save our - return PC. Save the signal mask if requested with a tail-call - for simplicity; it always returns zero. */ - ta ST_FLUSH_WINDOWS -#ifdef PIC - mov %o7,%g1 -2: - call 1f - nop -1: - sethi %hi(_GLOBAL_OFFSET_TABLE_-(2b-.)),%g2 - or %g2,%lo(_GLOBAL_OFFSET_TABLE_-(2b-.)),%g2 - add %g2,%o7,%g2 - sethi %hi(C_SYMBOL_NAME (__sigjmp_save)), %g3 - or %g3,%lo(C_SYMBOL_NAME (__sigjmp_save)), %g3 - st %sp, [%o0 + (JB_SP * 4)] - st %fp, [%o0 + (JB_FP * 4)] - mov %g1,%o7 - ld [%g2+%g3],%g1 - jmp %g1 - st %o7, [%o0+(JB_PC*4)] -#else - sethi %hi(C_SYMBOL_NAME (__sigjmp_save)), %g1 - st %sp, [%o0 + (JB_SP*4)] - or %lo(C_SYMBOL_NAME (__sigjmp_save)), %g1, %g1 - st %fp, [%o0 + (JB_FP*4)] - jmp %g1 - st %o7, [%o0 + (JB_PC*4)] -#endif /* PIC */ diff --git a/sysdeps/sparc/sparc32/Dist b/sysdeps/sparc/sparc32/Dist new file mode 100644 index 0000000000..6c3b23baa5 --- /dev/null +++ b/sysdeps/sparc/sparc32/Dist @@ -0,0 +1,4 @@ +dotmul.S umul.S +divrem.m4 sdiv.S udiv.S rem.S urem.S +alloca.S +sys/trap.h diff --git a/sysdeps/sparc/sparc32/Implies b/sysdeps/sparc/sparc32/Implies new file mode 100644 index 0000000000..39a34c5f57 --- /dev/null +++ b/sysdeps/sparc/sparc32/Implies @@ -0,0 +1 @@ +wordsize-32 diff --git a/sysdeps/sparc/sparc32/Makefile b/sysdeps/sparc/sparc32/Makefile new file mode 100644 index 0000000000..deec2f8145 --- /dev/null +++ b/sysdeps/sparc/sparc32/Makefile @@ -0,0 +1,51 @@ +# Copyright (C) 1991, 92, 93, 94, 95, 96, 97 Free Software Foundation, Inc. +# This file is part of the GNU C Library. + +# The GNU C Library is free software; you can redistribute it and/or +# modify it under the terms of the GNU Library General Public License +# as published by the Free Software Foundation; either version 2 of +# the License, or (at your option) any later version. + +# The GNU C Library is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +# Library General Public License for more details. + +# You should have received a copy of the GNU Library General Public +# License along with the GNU C Library; see the file COPYING.LIB. If not, +# write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, +# Boston, MA 02111-1307, USA. + +ifeq ($(subdir),gnulib) +routines = dotmul umul $(divrem) alloca +endif # gnulib + +# We distribute these files, even though they are generated, +# so as to avoid the need for a functioning m4 to build the library. +divrem := sdiv udiv rem urem + ++divrem-NAME-sdiv := div ++divrem-NAME-udiv := udiv ++divrem-NAME-rem := rem ++divrem-NAME-urem := urem ++divrem-NAME = $(+divrem-NAME-$(basename $(notdir $@))) ++divrem-OP-div := div ++divrem-OP-udiv := div ++divrem-OP-rem := rem ++divrem-OP-urem := rem ++divrem-S-div := true ++divrem-S-rem := true ++divrem-S-udiv := false ++divrem-S-urem := false +$(divrem:%=$(sysdep_dir)/sparc/%.S): $(sysdep_dir)/sparc/divrem.m4 + (echo "define(NAME,\`.$(+divrem-NAME)')\ + define(OP,\`$(+divrem-OP-$(+divrem-NAME))')\ + define(S,\`$(+divrem-S-$(+divrem-NAME))')\ + /* This file is generated from divrem.m4; DO NOT EDIT! */"; \ + cat $<) | $(M4) > $@-tmp +# Make it unwritable so noone will edit it by mistake. + -chmod a-w $@-tmp + mv -f $@-tmp $@ + test ! -d CVS || cvs commit -m'Regenerated from $<' $@ + +sysdep-realclean := $(sysdep-realclean) $(divrem:%=sysdeps/sparc/%.S) diff --git a/sysdeps/sparc/sparc32/__longjmp.S b/sysdeps/sparc/sparc32/__longjmp.S new file mode 100644 index 0000000000..36e1c170dd --- /dev/null +++ b/sysdeps/sparc/sparc32/__longjmp.S @@ -0,0 +1,59 @@ +/* Copyright (C) 1991, 1993, 1996, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +#define _ASM 1 +#include +#define ENV(reg) [%g1 + (reg * 4)] + +ENTRY (__longjmp) + /* Store our arguments in global registers so we can still + use them while unwinding frames and their register windows. */ + mov %o0, %g1 /* ENV in %g1 */ + orcc %o1, %g0, %g6 /* VAL in %g6 */ + be,a 0f /* Branch if zero; else skip delay slot. */ + mov 1, %g6 /* Delay slot only hit if zero: VAL = 1. */ +0: + + /* Cache target FP in register %g7. */ + ld ENV (JB_FP), %g7 + + /* Now we will loop, unwinding the register windows up the stack + until the restored %fp value matches the target value in %g7. */ + +loop: cmp %fp, %g7 /* Have we reached the target frame? */ + bl,a loop /* Loop while current fp is below target. */ + restore /* Unwind register window in delay slot. */ + be,a found /* Better have hit it exactly. */ + ld ENV (JB_SP), %o0 /* Delay slot: extract target SP. */ + +bogus: /* Get here only if the jmp_buf or stack is clobbered. */ + call C_SYMBOL_NAME (abort) + nop + unimp 0 + +found: /* We have unwound register windows so %fp matches the target. */ + cmp %o0, %sp /* Check jmp_buf SP vs register window. */ + bge,a sp_ok /* Saved must not be deeper than register. */ + mov %o0, %sp /* OK, install new SP. */ + b,a bogus /* Bogus, we lose. */ + +sp_ok: ld ENV (JB_PC), %o0 /* Extract target return PC. */ + jmp %o0 + 8 /* Return there. */ + mov %g6, %o0 /* Delay slot: set return value. */ diff --git a/sysdeps/sparc/sparc32/add_n.S b/sysdeps/sparc/sparc32/add_n.S new file mode 100644 index 0000000000..9852c256aa --- /dev/null +++ b/sysdeps/sparc/sparc32/add_n.S @@ -0,0 +1,226 @@ +! SPARC __mpn_add_n -- Add two limb vectors of the same length > 0 and store +! sum in a third limb vector. + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +#define res_ptr %o0 +#define s1_ptr %o1 +#define s2_ptr %o2 +#define size %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_add_n) +C_SYMBOL_NAME(__mpn_add_n): + xor s2_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L1 ! branch if alignment differs + nop +! ** V1a ** +L0: andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 + be L_v1 ! if no, branch + nop +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + addcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr +L_v1: addx %g0,%g0,%o4 ! save cy in register + cmp size,2 ! if size < 2 ... + bl Lend2 ! ... branch to tail code + subcc %g0,%o4,%g0 ! restore cy + + ld [s1_ptr+0],%g4 + addcc size,-10,size + ld [s1_ptr+4],%g1 + ldd [s2_ptr+0],%g2 + blt Lfin1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop1: addxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + addxcc %g4,%g2,%o4 + ld [s1_ptr+16],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+20],%g1 + ldd [s2_ptr+16],%g2 + std %o4,[res_ptr+8] + addxcc %g4,%g2,%o4 + ld [s1_ptr+24],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+28],%g1 + ldd [s2_ptr+24],%g2 + std %o4,[res_ptr+16] + addxcc %g4,%g2,%o4 + ld [s1_ptr+32],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+36],%g1 + ldd [s2_ptr+32],%g2 + std %o4,[res_ptr+24] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop1 + subcc %g0,%o4,%g0 ! restore cy + +Lfin1: addcc size,8-2,size + blt Lend1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1: addxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + addxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope1 + subcc %g0,%o4,%g0 ! restore cy +Lend1: addxcc %g4,%g2,%o4 + addxcc %g1,%g3,%o5 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + + andcc size,1,%g0 + be Lret1 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ + ld [s1_ptr+8],%g4 + ld [s2_ptr+8],%g2 + addxcc %g4,%g2,%o4 + st %o4,[res_ptr+8] + +Lret1: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb + +L1: xor s1_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L2 + nop +! ** V1b ** + mov s2_ptr,%g1 + mov s1_ptr,s2_ptr + b L0 + mov %g1,s1_ptr + +! ** V2 ** +/* If we come here, the alignment of s1_ptr and res_ptr as well as the + alignment of s2_ptr and res_ptr differ. Since there are only two ways + things can be aligned (that we care about) we now know that the alignment + of s1_ptr and s2_ptr are the same. */ + +L2: cmp size,1 + be Ljone + nop + andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 + be L_v2 ! if no, branch + nop +/* Add least significant limb separately to align s1_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + addcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr + +L_v2: addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + blt Lfin2 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + ldd [s1_ptr+8],%g2 + ldd [s2_ptr+8],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+8] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+12] + ldd [s1_ptr+16],%g2 + ldd [s2_ptr+16],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+16] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+20] + ldd [s1_ptr+24],%g2 + ldd [s2_ptr+24],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+24] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+28] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop2 + subcc %g0,%o4,%g0 ! restore cy + +Lfin2: addcc size,8-2,size + blt Lend2 + subcc %g0,%o4,%g0 ! restore cy +Loope2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + addxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + addxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope2 + subcc %g0,%o4,%g0 ! restore cy +Lend2: andcc size,1,%g0 + be Lret2 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ +Ljone: ld [s1_ptr],%g4 + ld [s2_ptr],%g2 + addxcc %g4,%g2,%o4 + st %o4,[res_ptr] + +Lret2: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb diff --git a/sysdeps/sparc/sparc32/addmul_1.S b/sysdeps/sparc/sparc32/addmul_1.S new file mode 100644 index 0000000000..375d25db6b --- /dev/null +++ b/sysdeps/sparc/sparc32/addmul_1.S @@ -0,0 +1,147 @@ +! SPARC __mpn_addmul_1 -- Multiply a limb vector with a limb and add +! the result to a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_addmul_1) +C_SYMBOL_NAME(__mpn_addmul_1): + ! Make S1_PTR and RES_PTR point at the end of their blocks + ! and put (- 4 x SIZE) in index/loop counter. + sll %o2,2,%o2 + add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval + add %o1,%o2,%o1 + sub %g0,%o2,%o2 + + cmp %o3,0xfff + bgu Large + nop + + ld [%o1+%o2],%o5 + mov 0,%o0 + b L0 + add %o4,-4,%o4 +Loop0: + addcc %o5,%g1,%g1 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g1,[%o4+%o2] +L0: wr %g0,%o3,%y + sra %o5,31,%g2 + and %o3,%g2,%g2 + andcc %g1,0,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,0,%g1 + sra %g1,20,%g4 + sll %g1,12,%g1 + rd %y,%g3 + srl %g3,20,%g3 + or %g1,%g3,%g1 + + addcc %g1,%o0,%g1 + addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne Loop0 + ld [%o4+%o2],%o5 + + addcc %o5,%g1,%g1 + addx %o0,%g0,%o0 + retl + st %g1,[%o4+%o2] + + +Large: ld [%o1+%o2],%o5 + mov 0,%o0 + sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 + b L1 + add %o4,-4,%o4 +Loop: + addcc %o5,%g3,%g3 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g3,[%o4+%o2] +L1: wr %g0,%o5,%y + and %o5,%g4,%g2 + andcc %g0,%g0,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%g0,%g1 + rd %y,%g3 + addcc %g3,%o0,%g3 + addx %g2,%g1,%o0 + addcc %o2,4,%o2 + bne Loop + ld [%o4+%o2],%o5 + + addcc %o5,%g3,%g3 + addx %o0,%g0,%o0 + retl + st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/sparc32/alloca.S b/sysdeps/sparc/sparc32/alloca.S new file mode 100644 index 0000000000..dcbd171163 --- /dev/null +++ b/sysdeps/sparc/sparc32/alloca.S @@ -0,0 +1,32 @@ +/* Copyright (C) 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "sysdep.h" + +/* Code produced by Sun's C compiler calls this function with two extra + arguments which it makes relocatable symbols but seem always to be + the constant 96; I have no idea what they are for. */ + +#ifndef NO_UNDERSCORES +#define __builtin_alloca ___builtin_alloca +#endif + +ENTRY (__builtin_alloca) + sub %sp, %o0, %sp /* Push some stack space. */ + retl /* Return; the returned buffer leaves 96 */ + add %sp, 96, %o0 /* bytes of register save area at the top. */ diff --git a/sysdeps/sparc/sparc32/bits/endian.h b/sysdeps/sparc/sparc32/bits/endian.h new file mode 100644 index 0000000000..f1a75c0652 --- /dev/null +++ b/sysdeps/sparc/sparc32/bits/endian.h @@ -0,0 +1,3 @@ +/* SPARC is big-endian. */ + +#define __BYTE_ORDER __BIG_ENDIAN diff --git a/sysdeps/sparc/sparc32/bits/setjmp.h b/sysdeps/sparc/sparc32/bits/setjmp.h new file mode 100644 index 0000000000..43bae1a496 --- /dev/null +++ b/sysdeps/sparc/sparc32/bits/setjmp.h @@ -0,0 +1,16 @@ +/* Define the machine-dependent type `jmp_buf'. SPARC version. */ + +#if defined (__USE_MISC) || defined (_ASM) +#define JB_SP 0 +#define JB_FP 1 +#define JB_PC 2 +#endif + +#ifndef _ASM +typedef int __jmp_buf[3]; +#endif + +/* Test if longjmp to JMPBUF would unwind the frame + containing a local variable at ADDRESS. */ +#define _JMPBUF_UNWINDS(jmpbuf, address) \ + ((int) (address) < (jmpbuf)[JB_SP]) diff --git a/sysdeps/sparc/sparc32/bsd-_setjmp.S b/sysdeps/sparc/sparc32/bsd-_setjmp.S new file mode 100644 index 0000000000..bf49765f4d --- /dev/null +++ b/sysdeps/sparc/sparc32/bsd-_setjmp.S @@ -0,0 +1,40 @@ +/* BSD `_setjmp' entry point to `sigsetjmp (..., 0)'. Sparc version. + Copyright (C) 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +ENTRY (_setjmp) + +#ifdef PIC + save %sp, -64, %sp +1: call 2f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 +2: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 + add %g1, %o7, %g1 + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g2 + restore + or %g2, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g2 + ld [%g1+%g2], %g1 +#else + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g1 + or %g1, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g1 +#endif + + jmp %g1 + mov %g0, %o1 /* Pass second argument of zero. */ diff --git a/sysdeps/sparc/sparc32/bsd-setjmp.S b/sysdeps/sparc/sparc32/bsd-setjmp.S new file mode 100644 index 0000000000..16dc260765 --- /dev/null +++ b/sysdeps/sparc/sparc32/bsd-setjmp.S @@ -0,0 +1,40 @@ +/* BSD `setjmp' entry point to `sigsetjmp (..., 1)'. Sparc version. + Copyright (C) 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + +ENTRY (setjmp) + +#ifdef PIC + save %sp, -64, %sp +1: call 2f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 +2: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %g1 + add %g1, %o7, %g1 + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g2 + restore + or %g2, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g2 + ld [%g1+%g2], %g1 +#else + sethi %hi(C_SYMBOL_NAME(__sigsetjmp)), %g1 + or %g1, %lo(C_SYMBOL_NAME(__sigsetjmp)), %g1 +#endif + + jmp %g1 + mov 1, %o1 /* Pass second argument of one. */ diff --git a/sysdeps/sparc/sparc32/divrem.m4 b/sysdeps/sparc/sparc32/divrem.m4 new file mode 100644 index 0000000000..665abf11ae --- /dev/null +++ b/sysdeps/sparc/sparc32/divrem.m4 @@ -0,0 +1,238 @@ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * NAME name of function to generate + * OP OP=div => %o0 / %o1; OP=rem => %o0 % %o1 + * S S=true => signed; S=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top `decade' of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + +define(N, `4')dnl +define(WORDSIZE, `32')dnl +define(TOPBITS, eval(WORDSIZE - N*((WORDSIZE-1)/N)))dnl +dnl +define(dividend, `%o0')dnl +define(divisor, `%o1')dnl +define(Q, `%o2')dnl +define(R, `%o3')dnl +define(ITER, `%o4')dnl +define(V, `%o5')dnl +dnl +dnl m4 reminder: ifelse(a,b,c,d) => if a is b, then c, else d +define(T, `%g1')dnl +define(SC, `%g7')dnl +ifelse(S, `true', `define(SIGN, `%g6')')dnl + +dnl +dnl This is the recursive definition for developing quotient digits. +dnl +dnl Parameters: +dnl $1 the current depth, 1 <= $1 <= N +dnl $2 the current accumulation of quotient bits +dnl N max depth +dnl +dnl We add a new bit to $2 and either recurse or insert the bits in +dnl the quotient. R, Q, and V are inputs and outputs as defined above; +dnl the condition codes are expected to reflect the input R, and are +dnl modified to reflect the output R. +dnl +define(DEVELOP_QUOTIENT_BITS, +` ! depth $1, accumulated bits $2 + bl L.$1.eval(2**N+$2) + srl V,1,V + ! remainder is positive + subcc R,V,R + ifelse($1, N, + ` b 9f + add Q, ($2*2+1), Q + ', ` DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2+1)')') +L.$1.eval(2**N+$2): + ! remainder is negative + addcc R,V,R + ifelse($1, N, + ` b 9f + add Q, ($2*2-1), Q + ', ` DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2-1)')') + ifelse($1, 1, `9:')')dnl + +#include "sysdep.h" +#ifdef __linux__ +#include +#else +#ifdef __svr4__ +#include +#else +#include +#endif +#endif + +FUNC(NAME) +ifelse(S, `true', +` ! compute sign of result; if neither is negative, no problem + orcc divisor, dividend, %g0 ! either negative? + bge 2f ! no, go do the divide +ifelse(OP, `div', +` xor divisor, dividend, SIGN ! compute sign in any case', +` mov dividend, SIGN ! sign of remainder matches dividend') + tst divisor + bge 1f + tst dividend + ! divisor is definitely negative; dividend might also be negative + bge 2f ! if dividend not negative... + sub %g0, divisor, divisor ! in any case, make divisor nonneg +1: ! dividend is negative, divisor is nonnegative + sub %g0, dividend, dividend ! make dividend nonnegative +2: +') + ! Ready to divide. Compute size of quotient; scale comparand. + orcc divisor, %g0, V + bne 1f + mov dividend, R + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp R, V ! if divisor exceeds dividend, done + blu Lgot_result ! (and algorithm fails otherwise) + clr Q + sethi %hi(1 << (WORDSIZE - TOPBITS - 1)), T + cmp R, T + blu Lnot_really_big + clr ITER + + ! `Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R.' + 1: + cmp V, T + bgeu 3f + mov 1, SC + sll V, N, V + b 1b + add ITER, 1, ITER + + ! Now compute SC. + 2: addcc V, V, V + bcc Lnot_too_big + add SC, 1, SC + + ! We get here if the divisor overflowed while shifting. + ! This means that R has the high-order bit set. + ! Restore V and subtract from R. + sll T, TOPBITS, T ! high order bit + srl V, 1, V ! rest of V + add V, T, V + b Ldo_single_div + sub SC, 1, SC + + Lnot_too_big: + 3: cmp V, R + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! V > R: went too far: back up 1 step + ! srl V, 1, V + ! dec SC + ! do single-bit divide steps + ! + ! We have to be careful here. We know that R >= V, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if R >= 0. Because both R and V may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc SC, 1, SC + bl Lend_regular_divide + nop + sub R, V, R + mov 1, Q + b Lend_single_divloop + nop + Lsingle_divloop: + sll Q, 1, Q + bl 1f + srl V, 1, V + ! R >= 0 + sub R, V, R + b 2f + add Q, 1, Q + 1: ! R < 0 + add R, V, R + sub Q, 1, Q + 2: + Lend_single_divloop: + subcc SC, 1, SC + bge Lsingle_divloop + tst R + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll V, N, V + cmp V, R + bleu 1b + addcc ITER, 1, ITER + be Lgot_result + sub ITER, 1, ITER + + tst R ! set up for initial iteration +Ldivloop: + sll Q, N, Q + DEVELOP_QUOTIENT_BITS(1, 0) +Lend_regular_divide: + subcc ITER, 1, ITER + bge Ldivloop + tst R + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) +ifelse(OP, `div', +` sub Q, 1, Q +', ` add R, divisor, R +') + +Lgot_result: +ifelse(S, `true', +` ! check to see if answer should be < 0 + tst SIGN + bl,a 1f + ifelse(OP, `div', `sub %g0, Q, Q', `sub %g0, R, R') +1:') + retl + ifelse(OP, `div', `mov Q, %o0', `mov R, %o0') diff --git a/sysdeps/sparc/sparc32/dl-machine.h b/sysdeps/sparc/sparc32/dl-machine.h new file mode 100644 index 0000000000..f7bcba0c2f --- /dev/null +++ b/sysdeps/sparc/sparc32/dl-machine.h @@ -0,0 +1,363 @@ +/* Machine-dependent ELF dynamic relocation inline functions. SPARC version. + Copyright (C) 1996, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, write to the Free Software Foundation, Inc., + 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + +#define ELF_MACHINE_NAME "sparc" + +#include +#include +#include +#include + + +/* Some SPARC opcodes we need to use for self-modifying code. */ +#define OPCODE_NOP 0x01000000 /* nop */ +#define OPCODE_CALL 0x40000000 /* call ?; add PC-rel word address */ +#define OPCODE_SETHI_G1 0x03000000 /* sethi ?, %g1; add value>>10 */ +#define OPCODE_JMP_G1 0x81c06000 /* jmp %g1+?; add lo 10 bits of value */ +#define OPCODE_SAVE_SP64 0x9de3bfc0 /* save %sp, -64, %sp */ + + +/* Return nonzero iff E_MACHINE is compatible with the running host. */ +static inline int +elf_machine_matches_host (Elf32_Half e_machine) +{ + return e_machine == EM_SPARC; +} + + +/* Return the link-time address of _DYNAMIC. Conveniently, this is the + first element of the GOT. This must be inlined in a function which + uses global data. */ +static inline Elf32_Addr +elf_machine_dynamic (void) +{ + register Elf32_Addr *got asm ("%l7"); + return *got; +} + +/* Return the run-time load address of the shared object. */ +static inline Elf32_Addr +elf_machine_load_address (void) +{ + register Elf32_Addr pc __asm("%o7"), got; + + /* Utilize the fact that a local .got entry will be partially + initialized at startup awaiting its RELATIVE fixup. */ + + __asm("sethi %%hi(.Load_address),%1\n" + ".Load_address:\n\t" + "call 1f\n\t" + "or %1,%%lo(.Load_address),%1\n" + "1:\tld [%%l7+%1],%1" + : "=r"(pc), "=r"(got)); + + return pc - got; +} + + Elf32_Addr addr; + + asm ( + "add %%fp,0x44,%%o2\n\t" /* o2 = point to argc */ + "ld [%%o2 - 4],%%o0\n\t" /* o0 = load argc */ + "sll %%o0, 2, %%o0\n\t" /* o0 = argc * sizeof (int) */ + "add %%o2,%%o0,%%o2\n\t" /* o2 = skip over argv */ + "add %%o2,4,%%o2\n\t" /* skip over null after argv */ + + /* Now %o2 is pointing to env, skip over that as well. */ + "1:\n\t" + "ld [%%o2],%%o0\n\t" + "cmp %%o0,0\n\t" + "bnz 1b\n\t" + "add %%o2,4,%%o2\n\t" + + /* Note that above, we want to advance the NULL after envp so + we always add 4. */ + + /* Now, search for the AT_BASE property. */ + "2:\n\t" + "ld [%%o2],%%o0\n\t" + "cmp %%o0,0\n\t" + "be,a 3f\n\t" + "or %%g0,%%g0,%0\n\t" + "cmp %%o0,7\n\t" /* AT_BASE = 7 */ + "be,a 3f\n\t" + "ld [%%o2+4],%0\n\t" + "b 2b\n\t" + "add %%o2,8,%%o2\n\t" + /* At this point %0 has the load address for the interpreter */ + "3:\n\t" + : "=r" (addr) + : /* no inputs */ + : "o0", "o2"); + return addr; +} + +#ifdef RESOLVE +/* Perform the relocation specified by RELOC and SYM (which is fully resolved). + MAP is the object containing the reloc. */ + +static inline void +elf_machine_rela (struct link_map *map, const Elf32_Rela *reloc, + const Elf32_Sym *sym, const struct r_found_version *version, + Elf32_Addr *const reloc_addr) +{ + Elf32_Addr loadbase; + + if (ELF32_R_TYPE (reloc->r_info) == R_SPARC_RELATIVE) + { +#ifndef RTLD_BOOTSTRAP + if (map != &_dl_rtld_map) /* Already done in rtld itself. */ +#endif + *reloc_addr += map->l_addr + reloc->r_addend; + } + else + { + const Elf32_Sym *const refsym = sym; + Elf32_Addr value; + if (sym->st_shndx != SHN_UNDEF && + ELF32_ST_BIND (sym->st_info) == STB_LOCAL) + value = map->l_addr; + else + { + value = RESOLVE (&sym, version, ELF32_R_TYPE (reloc->r_info)); + if (sym) + value += sym->st_value; + } + value += reloc->r_addend; /* Assume copy relocs have zero addend. */ + + switch (ELF32_R_TYPE (reloc->r_info)) + { + case R_SPARC_COPY: + if (sym->st_size > refsym->st_size + || (_dl_verbose && sym->st_size < refsym->st_size)) + { + extern char **_dl_argv; + const char *strtab; + + strtab = ((void *) map->l_addr + + map->l_info[DT_STRTAB]->d_un.d_ptr); + _dl_sysdep_error (_dl_argv[0] ?: "", + ": Symbol `", strtab + refsym->st_name, + "' has different size in shared object, " + "consider re-linking\n", NULL); + } + memcpy (reloc_addr, (void *) value, MIN (sym->st_size, + refsym->st_size)); + break; + case R_SPARC_GLOB_DAT: + case R_SPARC_32: + *reloc_addr = value; + break; + case R_SPARC_JMP_SLOT: + reloc_addr[1] = OPCODE_SETHI_G1 | (value >> 10); + reloc_addr[2] = OPCODE_JMP_G1 | (value & 0x3ff); + break; + case R_SPARC_8: + *(char *) reloc_addr = value; + break; + case R_SPARC_16: + *(short *) reloc_addr = value; + break; + case R_SPARC_DISP8: + *(char *) reloc_addr = (value - (Elf32_Addr) reloc_addr); + break; + case R_SPARC_DISP16: + *(short *) reloc_addr = (value - (Elf32_Addr) reloc_addr); + break; + case R_SPARC_DISP32: + *reloc_addr = (value - (Elf32_Addr) reloc_addr); + break; + case R_SPARC_LO10: + *reloc_addr = (*reloc_addr & ~0x3ff) | (value & 0x3ff); + break; + case R_SPARC_WDISP30: + *reloc_addr = ((*reloc_addr & 0xc0000000) + | ((value - (unsigned int) reloc_addr) >> 2)); + break; + case R_SPARC_HI22: + *reloc_addr = (*reloc_addr & 0xffc00000) | (value >> 10); + break; + case R_SPARC_NONE: /* Alright, Wilbur. */ + break; + default: + assert (! "unexpected dynamic reloc type"); + break; + } + } +} + +static inline void +elf_machine_lazy_rel (struct link_map *map, const Elf32_Rela *reloc) +{ + switch (ELF32_R_TYPE (reloc->r_info)) + { + case R_SPARC_NONE: + break; + case R_SPARC_JMP_SLOT: + break; + default: + assert (! "unexpected PLT reloc type"); + break; + } +} + +#endif /* RESOLVE */ + +/* Nonzero iff TYPE should not be allowed to resolve to one of + the main executable's symbols, as for a COPY reloc. */ +#define elf_machine_lookup_noexec_p(type) ((type) == R_SPARC_COPY) + +/* Nonzero iff TYPE describes relocation of a PLT entry, so + PLT entries should not be allowed to define the value. */ +#define elf_machine_lookup_noplt_p(type) ((type) == R_SPARC_JMP_SLOT) + +/* A reloc type used for ld.so cmdline arg lookups to reject PLT entries. */ +#define ELF_MACHINE_RELOC_NOPLT R_SPARC_JMP_SLOT + +/* The SPARC never uses Elf32_Rel relocations. */ +#define ELF_MACHINE_NO_REL 1 + +/* The SPARC overlaps DT_RELA and DT_PLTREL. */ +#define ELF_MACHINE_PLTREL_OVERLAP 1 + +/* Set up the loaded object described by L so its unrelocated PLT + entries will jump to the on-demand fixup code in dl-runtime.c. */ + +static inline int +elf_machine_runtime_setup (struct link_map *l, int lazy, int profile) +{ + Elf32_Addr *plt; + extern void _dl_runtime_resolve (Elf32_Word); + + if (l->l_info[DT_JMPREL] && lazy) + { + /* The entries for functions in the PLT have not yet been filled in. + Their initial contents will arrange when called to set the high 22 + bits of %g1 with an offset into the .rela.plt section and jump to + the beginning of the PLT. */ + plt = (Elf32_Addr *) (l->l_addr + l->l_info[DT_PLTGOT]->d_un.d_ptr); + + /* The beginning of the PLT does: + + save %sp, -64, %sp + pltpc: call _dl_runtime_resolve + nop + .word MAP + + This saves the register window containing the arguments, and the + PC value (pltpc) implicitly saved in %o7 by the call points near the + location where we store the link_map pointer for this object. */ + + plt[0] = OPCODE_SAVE_SP64; /* save %sp, -64, %sp */ + /* Construct PC-relative word address. */ + plt[1] = OPCODE_CALL | (((Elf32_Addr) &_dl_runtime_resolve - + (Elf32_Addr) &plt[1]) >> 2); + plt[2] = OPCODE_NOP; /* Fill call delay slot. */ + plt[3] = (Elf32_Addr *) l; + } + + return lazy; +} + +/* This code is used in dl-runtime.c to call the `fixup' function + and then redirect to the address it returns. */ +#define ELF_MACHINE_RUNTIME_TRAMPOLINE asm ("\ +# Trampoline for _dl_runtime_resolver + .globl _dl_runtime_resolve + .type _dl_runtime_resolve, @function +_dl_runtime_resolve: + t 1 + #call %g0 + # Pass two args to fixup: the PLT address computed from the PC saved + # in the PLT's call insn, and the reloc offset passed in %g1. + #ld [%o7 + 8], %o1 | Second arg, loaded from PLTPC[2]. + #call fixup + #shrl %g1, 22, %o0 | First arg, set in delay slot of call. + # Jump to the real function. + #jmpl %o0, %g0 + # In the delay slot of that jump, restore the register window + # saved by the first insn of the PLT. + #restore + .size _dl_runtime_resolve, . - _dl_runtime_resolve +"); + +/* The PLT uses Elf32_Rela relocs. */ +#define elf_machine_relplt elf_machine_rela + + +/* Mask identifying addresses reserved for the user program, + where the dynamic linker should not map anything. */ +#define ELF_MACHINE_USER_ADDRESS_MASK ??? + +/* Initial entry point code for the dynamic linker. + The C function `_dl_start' is the real entry point; + its return value is the user program's entry point. */ + +#define RTLD_START __asm__ ( \ +".text\n\ + .globl _start\n\ + .type _start,@function\n\ +_start:\n\ + /* Pass pointer to argument block to _dl_start. */\n\ + add %sp,64,%o0\n\ + call _dl_start\n\ + nop\n\ + \n\ + mov %o0,%l0\n\ + \n\ +2:\n\ + call 1f\n\ + nop\n\ +1:\n\ + sethi %hi(_GLOBAL_OFFSET_TABLE_-(2b-.)),%l2\n\ + sethi %hi(_dl_default_scope),%l3\n\ + or %l2,%lo(_GLOBAL_OFFSET_TABLE_-(2b-.)),%l2\n\ + or %l3,%lo(_dl_default_scope),%l3\n\ + add %o7,%l2,%l1\n\ + # %l1 has the GOT. %l3 has _dl_default_scope GOT offset\n\ + ld [%l1+%l3],%l4\n\ + # %l4 has pointer to _dl_default_scope. Now, load _dl_default_scope [2]\n\ + ld [%l4+8],%l4\n\ + # %l4 has _dl_default_scope [2]\n\ + # call _dl_init_next until it returns 0, pass _dl_default_scope [2]\n\ +3:\n\ + call _dl_init_next\n\ + mov %l4,%o0\n\ + cmp %o0,%g0\n\ + bz,a 4f\n\ + nop\n\ + call %o0\n\ + /* Pass pointer to argument block to this init function */\n\ + add %sp,64,%o0\n\ + b,a 3b\n\ +4:\n\ + # Clear the _dl_starting_up variable and pass _dl_fini in %g1 as per ELF ABI.\n\ + sethi %hi(_dl_starting_up),%l4\n\ + sethi %hi(_dl_fini),%l3\n\ + or %l4,%lo(_dl_starting_up),%l4\n\ + or %l3,%lo(_dl_fini),%l3\n\ + # clear _dl_starting_up\n\ + ld [%l1+%l4],%l5\n\ + st %g0,[%l5]\n\ + # load out fini function for atexit in %g1\n\ + ld [%l3+%l1],%g1\n\ + # jump to the user program entry point.\n\ + jmpl %l0,%g0\n\ + nop\n\ +"); diff --git a/sysdeps/sparc/sparc32/dotmul.S b/sysdeps/sparc/sparc32/dotmul.S new file mode 100644 index 0000000000..7ce695cb7a --- /dev/null +++ b/sysdeps/sparc/sparc32/dotmul.S @@ -0,0 +1,123 @@ +/* + * Signed multiply, from Appendix E of the Sparc Version 8 + * Architecture Manual. + */ + +/* + * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of + * the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. + */ + +#include "sysdep.h" +ENTRY(.mul) + mov %o0, %y ! multiplier -> Y + andncc %o0, 0xfff, %g0 ! test bits 12..31 + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + ! If %o0 was negative, the result is + ! (%o0 * %o1) + (%o1 << 32)) + ! We fix that here. + +#if 0 + tst %o0 + bge 1f + rd %y, %o0 + + ! %o0 was indeed negative; fix upper 32 bits of result by subtracting + ! %o1 (i.e., return %o4 - %o1 in %o1). + retl + sub %o4, %o1, %o1 + +1: + retl + mov %o4, %o1 +#else + /* Faster code adapted from tege@sics.se's code for umul.S. */ + sra %o0, 31, %o2 ! make mask from sign bit + and %o1, %o2, %o2 ! %o2 = 0 or %o1, depending on sign of %o0 + rd %y, %o0 ! get lower half of product + retl + sub %o4, %o2, %o1 ! subtract compensation + ! and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the low part of the + * result; %y has the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * --hi-- ----low-part---- + * + * The upper 12 bits of %o4 should be sign-extended to form the + * high part of the product (i.e., highpart = %o4 >> 20). + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20, zero fill at left + or %o5, %o0, %o0 ! construct low part of result + retl + sra %o4, 20, %o1 ! ... and extract high part of result diff --git a/sysdeps/sparc/sparc32/e_sqrt.c b/sysdeps/sparc/sparc32/e_sqrt.c new file mode 100644 index 0000000000..a98ae21ace --- /dev/null +++ b/sysdeps/sparc/sparc32/e_sqrt.c @@ -0,0 +1,34 @@ +/* Copyright (C) 1991, 1992, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include +#include + +#ifndef __GNUC__ + #error This file uses GNU C extensions; you must compile with GCC. +#endif + +/* Return the square root of X. */ +double +__ieee754_sqrt (x) + double x; +{ + register double result; + asm ("fsqrtd %1, %0" : "=f" (result) : "f" (x)); + return result; +} diff --git a/sysdeps/sparc/sparc32/elf/Makefile b/sysdeps/sparc/sparc32/elf/Makefile new file mode 100644 index 0000000000..319fbdef10 --- /dev/null +++ b/sysdeps/sparc/sparc32/elf/Makefile @@ -0,0 +1,4 @@ +# Sparc/ELF specific definitions. + +# The assembler on SPARC needs the -fPIC flag even when it's assembler code. +ASFLAGS-.so = -fPIC diff --git a/sysdeps/sparc/sparc32/elf/start.c b/sysdeps/sparc/sparc32/elf/start.c new file mode 100644 index 0000000000..f1e80195ca --- /dev/null +++ b/sysdeps/sparc/sparc32/elf/start.c @@ -0,0 +1,68 @@ +/* Copyright (C) 1991, 1992, 1993, 1994, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include +#include +#include +#include + +extern char **__environ; + +extern void __libc_init_first __P ((int argc, char **argv, char **envp)); +extern int main __P ((int argc, char **argv, char **envp)); + +register long int sp asm("%sp"), fp asm("%fp"); + +void +_start (void) +{ + /* It is important that these be declared `register'. + Otherwise, when compiled without optimization, they are put on the + stack, which loses completely after we zero the FP. */ + register int argc; + register char **argv, **envp; + register long int g1 asm ("%g1"); + unsigned long int copy_g1 = g1; + + /* Unwind the frame built when we entered the function. */ + asm("restore"); + if (copy_g1) + atexit (copy_g1); + + /* And clear the frame pointer. */ + fp = 0; + + /* The argument info starts after one register + window (64 bytes) past the SP. */ + argc = ((int *) sp)[16]; + argv = (char **) &((int *) sp)[17]; + envp = &argv[argc + 1]; + __environ = envp; + + /* Allocate 24 bytes of stack space for the register save area. */ + sp -= 24; + __libc_init_first (argc, argv, envp); +#ifdef ELF_INIT_FINI + { + extern void _fini (void); + _init (); + atexit (_fini); + } +#endif + exit (main (argc, argv, envp)); +} diff --git a/sysdeps/sparc/sparc32/fpu/bits/fenv.h b/sysdeps/sparc/sparc32/fpu/bits/fenv.h new file mode 100644 index 0000000000..0560dd99b8 --- /dev/null +++ b/sysdeps/sparc/sparc32/fpu/bits/fenv.h @@ -0,0 +1,76 @@ +/* Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _FENV_H +#error "Never use directly; include instead." +#endif + + +/* Define bits representing the exception. We use the bit positions + of the appropriate accrued exception bits from the FSR. */ +enum + { + FE_INVALID = (1 << 9), +#define FE_INVALID FE_INVALID + FE_OVERFLOW = (1 << 8), +#define FE_OVERFLOW FE_OVERFLOW + FE_UNDERFLOW = (1 << 7), +#define FE_UNDERFLOW FE_UNDERFLOW + FE_DIVBYZERO = (1 << 6), +#define FE_DIVBYZERO FE_DIVBYZERO + FE_INEXACT = (1 << 5) +#define FE_INEXACT FE_INEXACT + }; + +#define FE_ALL_EXCEPT \ + (FE_INEXACT | FE_DIVBYZERO | FE_UNDERFLOW | FE_OVERFLOW | FE_INVALID) + +/* The Sparc FPU supports all of the four defined rounding modes. We + use again the bit positions in the FPU control word as the values + for the appropriate macros. */ +enum + { + FE_TONEAREST = (0U << 30), +#define FE_TONEAREST FE_TONEAREST + FE_TOWARDSZERO = (1U << 30), +#define FE_TOWARDSZERO FE_TOWARDSZERO + FE_UPWARD = (2U << 30), +#define FE_UPWARD FE_UPWARD + FE_DOWNWARD = (3U << 30) +#define FE_DOWNWARD FE_DOWNWARD + }; + +#define __FE_ROUND_MASK (3U << 30) + +/* Type representing exception flags. */ +typedef unsigned int fexcept_t; + +/* Type representing floating-point environment. */ +typedef unsigned int fenv_t; + +/* If the default argument is used we use this value. */ +#define FE_DFL_ENV ((fenv_t *) -1) + +#ifdef __USE_GNU +/* Floating-point environment where none of the exception is masked. */ +# define FE_NOMASK_ENV ((fenv_t *) -2) +#endif + +/* For internal use only: access the fp state register. */ +#define __fenv_stfsr(X) __asm__("stfsr %0" : "=m"(X)) +#define __fenv_ldfsr(X) __asm__ __volatile__("ldfsr %0" : : "m"(X)) diff --git a/sysdeps/sparc/sparc32/fpu/fpu_control.h b/sysdeps/sparc/sparc32/fpu/fpu_control.h new file mode 100644 index 0000000000..5ef3824702 --- /dev/null +++ b/sysdeps/sparc/sparc32/fpu/fpu_control.h @@ -0,0 +1,69 @@ +/* FPU control word bits. SPARC version. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Miguel de Icaza + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _FPU_CONTROL_H +#define _FPU_CONTROL_H 1 + + +#include + +/* precision control */ +#define _FPU_EXTENDED 0x00000000 /* RECOMMENDED */ +#define _FPU_DOUBLE 0x20000000 +#define _FPU_80BIT 0x30000000 +#define _FPU_SINGLE 0x10000000 /* DO NOT USE */ + +/* rounding control / Sparc */ +#define _FPU_RC_DOWN 0xc0000000 +#define _FPU_RC_UP 0x80000000 +#define _FPU_RC_ZERO 0x40000000 +#define _FPU_RC_NEAREST 0x0 /* RECOMMENDED */ + +#define _FPU_RESERVED 0x30300000 /* Reserved bits in cw */ + + +/* Now two recommended cw */ + +/* Linux default: + - extended precision + - rounding to nearest + - exceptions on overflow, zero divide and NaN */ +#define _FPU_DEFAULT 0x1e + +/* IEEE: same as above, but exceptions */ +#define _FPU_IEEE 0x0 + +/* Type of the control word. */ +typedef unsigned int fpu_control_t; + +#define _FPU_GETCW(cw) __asm__ ("st %%fsr,%0" : "=m" (*&cw)) +#define _FPU_SETCW(cw) __asm__ ("ld %0,%%fsr" : : "m" (*&cw)) + +/* Default control word set at startup. */ +extern fpu_control_t __fpu_control; + +__BEGIN_DECLS + +/* Called at startup. It can be used to manipulate fpu control register. */ +extern void __setfpucw __P ((fpu_control_t)); + +__END_DECLS + +#endif /* fpu_control.h */ diff --git a/sysdeps/sparc/sparc32/lshift.S b/sysdeps/sparc/sparc32/lshift.S new file mode 100644 index 0000000000..4f0595f2fb --- /dev/null +++ b/sysdeps/sparc/sparc32/lshift.S @@ -0,0 +1,95 @@ +! sparc __mpn_lshift -- + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! src_ptr %o1 +! size %o2 +! cnt %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_lshift) +C_SYMBOL_NAME(__mpn_lshift): + sll %o2,2,%g1 + add %o1,%g1,%o1 ! make %o1 point at end of src + ld [%o1-4],%g2 ! load first limb + sub %g0,%o3,%o5 ! negate shift count + add %o0,%g1,%o0 ! make %o0 point at end of res + add %o2,-1,%o2 + andcc %o2,4-1,%g4 ! number of limbs in first loop + srl %g2,%o5,%g1 ! compute function result + be L0 ! if multiple of 4 limbs, skip first loop + st %g1,[%sp+80] + + sub %o2,%g4,%o2 ! adjust count for main loop + +Loop0: ld [%o1-8],%g3 + add %o0,-4,%o0 + add %o1,-4,%o1 + addcc %g4,-1,%g4 + sll %g2,%o3,%o4 + srl %g3,%o5,%g1 + mov %g3,%g2 + or %o4,%g1,%o4 + bne Loop0 + st %o4,[%o0+0] + +L0: tst %o2 + be Lend + nop + +Loop: ld [%o1-8],%g3 + add %o0,-16,%o0 + addcc %o2,-4,%o2 + sll %g2,%o3,%o4 + srl %g3,%o5,%g1 + + ld [%o1-12],%g2 + sll %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0+12] + srl %g2,%o5,%g1 + + ld [%o1-16],%g3 + sll %g2,%o3,%o4 + or %g4,%g1,%g4 + st %g4,[%o0+8] + srl %g3,%o5,%g1 + + ld [%o1-20],%g2 + sll %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0+4] + srl %g2,%o5,%g1 + + add %o1,-16,%o1 + or %g4,%g1,%g4 + bne Loop + st %g4,[%o0+0] + +Lend: sll %g2,%o3,%g2 + st %g2,[%o0-4] + retl + ld [%sp+80],%o0 diff --git a/sysdeps/sparc/sparc32/memcopy.h b/sysdeps/sparc/sparc32/memcopy.h new file mode 100644 index 0000000000..ff73f453da --- /dev/null +++ b/sysdeps/sparc/sparc32/memcopy.h @@ -0,0 +1,21 @@ +/* Copyright (C) 1991, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include +#undef reg_char +#define reg_char int diff --git a/sysdeps/sparc/sparc32/mul_1.S b/sysdeps/sparc/sparc32/mul_1.S new file mode 100644 index 0000000000..142fd8ba2a --- /dev/null +++ b/sysdeps/sparc/sparc32/mul_1.S @@ -0,0 +1,199 @@ +! SPARC __mpn_mul_1 -- Multiply a limb vector with a limb and store +! the result in a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +! ADD CODE FOR SMALL MULTIPLIERS! +!1: ld +! st +! +!2: ld ,a +! addxcc a,a,x +! st x, +! +!3_unrolled: +! ld ,a +! addxcc a,a,x1 ! 2a + cy +! addx %g0,%g0,x2 +! addcc a,x1,x ! 3a + c +! st x, +! +! ld ,a +! addxcc a,a,y1 +! addx %g0,%g0,y2 +! addcc a,y1,x +! st x, +! +!4_unrolled: +! ld ,a +! srl a,2,x1 ! 4a +! addxcc y2,x1,x +! sll a,30,x2 +! st x, +! +! ld ,a +! srl a,2,y1 +! addxcc x2,y1,y +! sll a,30,y2 +! st x, +! +!5_unrolled: +! ld ,a +! srl a,2,x1 ! 4a +! addxcc a,x1,x ! 5a + c +! sll a,30,x2 +! addx %g0,x2,x2 +! st x, +! +! ld ,a +! srl a,2,y1 +! addxcc a,y1,x +! sll a,30,y2 +! addx %g0,y2,y2 +! st x, +! +!8_unrolled: +! ld ,a +! srl a,3,x1 ! 8a +! addxcc y2,x1,x +! sll a,29,x2 +! st x, +! +! ld ,a +! srl a,3,y1 +! addxcc x2,y1,y +! sll a,29,y2 +! st x, + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_mul_1) +C_SYMBOL_NAME(__mpn_mul_1): + ! Make S1_PTR and RES_PTR point at the end of their blocks + ! and put (- 4 x SIZE) in index/loop counter. + sll %o2,2,%o2 + add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval + add %o1,%o2,%o1 + sub %g0,%o2,%o2 + + cmp %o3,0xfff + bgu Large + nop + + ld [%o1+%o2],%o5 + mov 0,%o0 + b L0 + add %o4,-4,%o4 +Loop0: + st %g1,[%o4+%o2] +L0: wr %g0,%o3,%y + sra %o5,31,%g2 + and %o3,%g2,%g2 + andcc %g1,0,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,0,%g1 + sra %g1,20,%g4 + sll %g1,12,%g1 + rd %y,%g3 + srl %g3,20,%g3 + or %g1,%g3,%g1 + + addcc %g1,%o0,%g1 + addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne,a Loop0 + ld [%o1+%o2],%o5 + + retl + st %g1,[%o4+%o2] + + +Large: ld [%o1+%o2],%o5 + mov 0,%o0 + sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 + b L1 + add %o4,-4,%o4 +Loop: + st %g3,[%o4+%o2] +L1: wr %g0,%o5,%y + and %o5,%g4,%g2 ! g2 = S1_LIMB iff S2_LIMB < 0, else 0 + andcc %g0,%g0,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%g0,%g1 + rd %y,%g3 + addcc %g3,%o0,%g3 + addx %g2,%g1,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne,a Loop + ld [%o1+%o2],%o5 + + retl + st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/sparc32/rem.S b/sysdeps/sparc/sparc32/rem.S new file mode 100644 index 0000000000..d50f1af630 --- /dev/null +++ b/sysdeps/sparc/sparc32/rem.S @@ -0,0 +1,369 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .rem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include +#else +#ifdef __svr4__ +#include +#else +#include +#endif +#endif + +ENTRY(.rem) + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + mov %o0, %g6 ! sign of remainder matches %o0 + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + + +Lgot_result: + ! check to see if answer should be < 0 + tst %g6 + bl,a 1f + sub %g0, %o3, %o3 +1: + retl + mov %o3, %o0 diff --git a/sysdeps/sparc/sparc32/rshift.S b/sysdeps/sparc/sparc32/rshift.S new file mode 100644 index 0000000000..fea4f3b926 --- /dev/null +++ b/sysdeps/sparc/sparc32/rshift.S @@ -0,0 +1,92 @@ +! sparc __mpn_rshift -- + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr %o0 +! src_ptr %o1 +! size %o2 +! cnt %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_rshift) +C_SYMBOL_NAME(__mpn_rshift): + ld [%o1],%g2 ! load first limb + sub %g0,%o3,%o5 ! negate shift count + add %o2,-1,%o2 + andcc %o2,4-1,%g4 ! number of limbs in first loop + sll %g2,%o5,%g1 ! compute function result + be L0 ! if multiple of 4 limbs, skip first loop + st %g1,[%sp+80] + + sub %o2,%g4,%o2 ! adjust count for main loop + +Loop0: ld [%o1+4],%g3 + add %o0,4,%o0 + add %o1,4,%o1 + addcc %g4,-1,%g4 + srl %g2,%o3,%o4 + sll %g3,%o5,%g1 + mov %g3,%g2 + or %o4,%g1,%o4 + bne Loop0 + st %o4,[%o0-4] + +L0: tst %o2 + be Lend + nop + +Loop: ld [%o1+4],%g3 + add %o0,16,%o0 + addcc %o2,-4,%o2 + srl %g2,%o3,%o4 + sll %g3,%o5,%g1 + + ld [%o1+8],%g2 + srl %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0-16] + sll %g2,%o5,%g1 + + ld [%o1+12],%g3 + srl %g2,%o3,%o4 + or %g4,%g1,%g4 + st %g4,[%o0-12] + sll %g3,%o5,%g1 + + ld [%o1+16],%g2 + srl %g3,%o3,%g4 + or %o4,%g1,%o4 + st %o4,[%o0-8] + sll %g2,%o5,%g1 + + add %o1,16,%o1 + or %g4,%g1,%g4 + bne Loop + st %g4,[%o0-4] + +Lend: srl %g2,%o3,%g2 + st %g2,[%o0-0] + retl + ld [%sp+80],%o0 diff --git a/sysdeps/sparc/sparc32/sdiv.S b/sysdeps/sparc/sparc32/sdiv.S new file mode 100644 index 0000000000..02ed2e973c --- /dev/null +++ b/sysdeps/sparc/sparc32/sdiv.S @@ -0,0 +1,369 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .div name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include +#else +#ifdef __svr4__ +#include +#else +#include +#endif +#endif + +ENTRY(.div) + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + xor %o1, %o0, %g6 ! compute sign in any case + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +Lgot_result: + ! check to see if answer should be < 0 + tst %g6 + bl,a 1f + sub %g0, %o2, %o2 +1: + retl + mov %o2, %o0 diff --git a/sysdeps/sparc/sparc32/setjmp.S b/sysdeps/sparc/sparc32/setjmp.S new file mode 100644 index 0000000000..772ec723a3 --- /dev/null +++ b/sysdeps/sparc/sparc32/setjmp.S @@ -0,0 +1,54 @@ +/* Copyright (C) 1991, 1993, 1994, 1996, 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include +#include + +#define _ASM 1 +#include + +ENTRY (__sigsetjmp) + /* Save our SP and FP; in the delay slot of the jump, save our + return PC. Save the signal mask if requested with a tail-call + for simplicity; it always returns zero. */ + ta ST_FLUSH_WINDOWS +#ifdef PIC + mov %o7,%g1 +2: + call 1f + nop +1: + sethi %hi(_GLOBAL_OFFSET_TABLE_-(2b-.)),%g2 + or %g2,%lo(_GLOBAL_OFFSET_TABLE_-(2b-.)),%g2 + add %g2,%o7,%g2 + sethi %hi(C_SYMBOL_NAME (__sigjmp_save)), %g3 + or %g3,%lo(C_SYMBOL_NAME (__sigjmp_save)), %g3 + st %sp, [%o0 + (JB_SP * 4)] + st %fp, [%o0 + (JB_FP * 4)] + mov %g1,%o7 + ld [%g2+%g3],%g1 + jmp %g1 + st %o7, [%o0+(JB_PC*4)] +#else + sethi %hi(C_SYMBOL_NAME (__sigjmp_save)), %g1 + st %sp, [%o0 + (JB_SP*4)] + or %lo(C_SYMBOL_NAME (__sigjmp_save)), %g1, %g1 + st %fp, [%o0 + (JB_FP*4)] + jmp %g1 + st %o7, [%o0 + (JB_PC*4)] +#endif /* PIC */ diff --git a/sysdeps/sparc/sparc32/sparcv8/addmul_1.S b/sysdeps/sparc/sparc32/sparcv8/addmul_1.S new file mode 100644 index 0000000000..fb9ea7cf0e --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/addmul_1.S @@ -0,0 +1,124 @@ +! SPARC v8 __mpn_addmul_1 -- Multiply a limb vector with a limb and +! add the result to a second limb vector. + +! Copyright (C) 1992, 1993, 1994, 1995 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_addmul_1) +C_SYMBOL_NAME(__mpn_addmul_1): + orcc %g0,%g0,%g2 + ld [%o1+0],%o4 ! 1 + + sll %o2,4,%g1 + and %g1,(4-1)<<4,%g1 +#if PIC + mov %o7,%g4 ! Save return address register + call 1f + add %o7,LL-1f,%g3 +1: mov %g4,%o7 ! Restore return address register +#else + sethi %hi(LL),%g3 + or %g3,%lo(LL),%g3 +#endif + jmp %g3+%g1 + nop +LL: +LL00: add %o0,-4,%o0 + b Loop00 /* 4, 8, 12, ... */ + add %o1,-4,%o1 + nop +LL01: b Loop01 /* 1, 5, 9, ... */ + nop + nop + nop +LL10: add %o0,-12,%o0 /* 2, 6, 10, ... */ + b Loop10 + add %o1,4,%o1 + nop +LL11: add %o0,-8,%o0 /* 3, 7, 11, ... */ + b Loop11 + add %o1,-8,%o1 + nop + +1: addcc %g3,%g2,%g3 ! 1 + ld [%o1+4],%o4 ! 2 + rd %y,%g2 ! 1 + addx %g0,%g2,%g2 + ld [%o0+0],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+0] ! 1 +Loop00: umul %o4,%o3,%g3 ! 2 + ld [%o0+4],%g1 ! 2 + addxcc %g3,%g2,%g3 ! 2 + ld [%o1+8],%o4 ! 3 + rd %y,%g2 ! 2 + addx %g0,%g2,%g2 + nop + addcc %g1,%g3,%g3 + st %g3,[%o0+4] ! 2 +Loop11: umul %o4,%o3,%g3 ! 3 + addxcc %g3,%g2,%g3 ! 3 + ld [%o1+12],%o4 ! 4 + rd %y,%g2 ! 3 + add %o1,16,%o1 + addx %g0,%g2,%g2 + ld [%o0+8],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+8] ! 3 +Loop10: umul %o4,%o3,%g3 ! 4 + addxcc %g3,%g2,%g3 ! 4 + ld [%o1+0],%o4 ! 1 + rd %y,%g2 ! 4 + addx %g0,%g2,%g2 + ld [%o0+12],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+12] ! 4 + add %o0,16,%o0 + addx %g0,%g2,%g2 +Loop01: addcc %o2,-4,%o2 + bg 1b + umul %o4,%o3,%g3 ! 1 + + addcc %g3,%g2,%g3 ! 4 + rd %y,%g2 ! 4 + addx %g0,%g2,%g2 + ld [%o0+0],%g1 ! 2 + addcc %g1,%g3,%g3 + st %g3,[%o0+0] ! 4 + addx %g0,%g2,%o0 + + retl + nop + + +! umul, ld, addxcc, rd, st + +! umul, ld, addxcc, rd, ld, addcc, st, addx + diff --git a/sysdeps/sparc/sparc32/sparcv8/mul_1.S b/sysdeps/sparc/sparc32/sparcv8/mul_1.S new file mode 100644 index 0000000000..b641feb453 --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/mul_1.S @@ -0,0 +1,99 @@ +! SPARC v8 __mpn_mul_1 -- Multiply a limb vector with a single limb and +! store the product in a second limb vector. + +! Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 8 + .global C_SYMBOL_NAME(__mpn_mul_1) +C_SYMBOL_NAME(__mpn_mul_1): + sll %o2,4,%g1 + and %g1,(4-1)<<4,%g1 +#if PIC + mov %o7,%g4 ! Save return address register + call 1f + add %o7,LL-1f,%g3 +1: mov %g4,%o7 ! Restore return address register +#else + sethi %hi(LL),%g3 + or %g3,%lo(LL),%g3 +#endif + jmp %g3+%g1 + ld [%o1+0],%o4 ! 1 +LL: +LL00: add %o0,-4,%o0 + add %o1,-4,%o1 + b Loop00 /* 4, 8, 12, ... */ + orcc %g0,%g0,%g2 +LL01: b Loop01 /* 1, 5, 9, ... */ + orcc %g0,%g0,%g2 + nop + nop +LL10: add %o0,-12,%o0 /* 2, 6, 10, ... */ + add %o1,4,%o1 + b Loop10 + orcc %g0,%g0,%g2 + nop +LL11: add %o0,-8,%o0 /* 3, 7, 11, ... */ + add %o1,-8,%o1 + b Loop11 + orcc %g0,%g0,%g2 + +Loop: addcc %g3,%g2,%g3 ! 1 + ld [%o1+4],%o4 ! 2 + st %g3,[%o0+0] ! 1 + rd %y,%g2 ! 1 +Loop00: umul %o4,%o3,%g3 ! 2 + addxcc %g3,%g2,%g3 ! 2 + ld [%o1+8],%o4 ! 3 + st %g3,[%o0+4] ! 2 + rd %y,%g2 ! 2 +Loop11: umul %o4,%o3,%g3 ! 3 + addxcc %g3,%g2,%g3 ! 3 + ld [%o1+12],%o4 ! 4 + add %o1,16,%o1 + st %g3,[%o0+8] ! 3 + rd %y,%g2 ! 3 +Loop10: umul %o4,%o3,%g3 ! 4 + addxcc %g3,%g2,%g3 ! 4 + ld [%o1+0],%o4 ! 1 + st %g3,[%o0+12] ! 4 + add %o0,16,%o0 + rd %y,%g2 ! 4 + addx %g0,%g2,%g2 +Loop01: addcc %o2,-4,%o2 + bg Loop + umul %o4,%o3,%g3 ! 1 + + addcc %g3,%g2,%g3 ! 4 + st %g3,[%o0+0] ! 4 + rd %y,%g2 ! 4 + + retl + addx %g0,%g2,%o0 diff --git a/sysdeps/sparc/sparc32/sparcv8/submul_1.S b/sysdeps/sparc/sparc32/sparcv8/submul_1.S new file mode 100644 index 0000000000..e40119d011 --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/submul_1.S @@ -0,0 +1,58 @@ +! SPARC v8 __mpn_submul_1 -- Multiply a limb vector with a limb and +! subtract the result from a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_submul_1) +C_SYMBOL_NAME(__mpn_submul_1): + sub %g0,%o2,%o2 ! negate ... + sll %o2,2,%o2 ! ... and scale size + sub %o1,%o2,%o1 ! o1 is offset s1_ptr + sub %o0,%o2,%g1 ! g1 is offset res_ptr + + mov 0,%o0 ! clear cy_limb + +Loop: ld [%o1+%o2],%o4 + ld [%g1+%o2],%g2 + umul %o4,%o3,%o5 + rd %y,%g3 + addcc %o5,%o0,%o5 + addx %g3,0,%o0 + subcc %g2,%o5,%g2 + addx %o0,0,%o0 + st %g2,[%g1+%o2] + + addcc %o2,4,%o2 + bne Loop + nop + + retl + nop diff --git a/sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S b/sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S new file mode 100644 index 0000000000..49c2398806 --- /dev/null +++ b/sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S @@ -0,0 +1,186 @@ +! SPARC __udiv_qrnnd division support, used from longlong.h. + +! Copyright (C) 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + + +! INPUT PARAMETERS +! rem_ptr o0 +! n1 o1 +! n0 o2 +! d o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__udiv_qrnnd) +C_SYMBOL_NAME(__udiv_qrnnd): + tst %o3 + bneg Largedivisor + mov 8,%g1 + + b Lp1 + addxcc %o2,%o2,%o2 + +Lplop: bcc Ln1 + addxcc %o2,%o2,%o2 +Lp1: addx %o1,%o1,%o1 + subcc %o1,%o3,%o4 + bcc Ln2 + addxcc %o2,%o2,%o2 +Lp2: addx %o1,%o1,%o1 + subcc %o1,%o3,%o4 + bcc Ln3 + addxcc %o2,%o2,%o2 +Lp3: addx %o1,%o1,%o1 + subcc %o1,%o3,%o4 + bcc Ln4 + addxcc %o2,%o2,%o2 +Lp4: addx %o1,%o1,%o1 + addcc %g1,-1,%g1 + bne Lplop + subcc %o1,%o3,%o4 + bcc Ln5 + addxcc %o2,%o2,%o2 +Lp5: st %o1,[%o0] + retl + xnor %g0,%o2,%o0 + +Lnlop: bcc Lp1 + addxcc %o2,%o2,%o2 +Ln1: addx %o4,%o4,%o4 + subcc %o4,%o3,%o1 + bcc Lp2 + addxcc %o2,%o2,%o2 +Ln2: addx %o4,%o4,%o4 + subcc %o4,%o3,%o1 + bcc Lp3 + addxcc %o2,%o2,%o2 +Ln3: addx %o4,%o4,%o4 + subcc %o4,%o3,%o1 + bcc Lp4 + addxcc %o2,%o2,%o2 +Ln4: addx %o4,%o4,%o4 + addcc %g1,-1,%g1 + bne Lnlop + subcc %o4,%o3,%o1 + bcc Lp5 + addxcc %o2,%o2,%o2 +Ln5: st %o4,[%o0] + retl + xnor %g0,%o2,%o0 + +Largedivisor: + and %o2,1,%o5 ! %o5 = n0 & 1 + + srl %o2,1,%o2 + sll %o1,31,%g2 + or %g2,%o2,%o2 ! %o2 = lo(n1n0 >> 1) + srl %o1,1,%o1 ! %o1 = hi(n1n0 >> 1) + + and %o3,1,%g2 + srl %o3,1,%g3 ! %g3 = floor(d / 2) + add %g3,%g2,%g3 ! %g3 = ceil(d / 2) + + b LLp1 + addxcc %o2,%o2,%o2 + +LLplop: bcc LLn1 + addxcc %o2,%o2,%o2 +LLp1: addx %o1,%o1,%o1 + subcc %o1,%g3,%o4 + bcc LLn2 + addxcc %o2,%o2,%o2 +LLp2: addx %o1,%o1,%o1 + subcc %o1,%g3,%o4 + bcc LLn3 + addxcc %o2,%o2,%o2 +LLp3: addx %o1,%o1,%o1 + subcc %o1,%g3,%o4 + bcc LLn4 + addxcc %o2,%o2,%o2 +LLp4: addx %o1,%o1,%o1 + addcc %g1,-1,%g1 + bne LLplop + subcc %o1,%g3,%o4 + bcc LLn5 + addxcc %o2,%o2,%o2 +LLp5: add %o1,%o1,%o1 ! << 1 + tst %g2 + bne Oddp + add %o5,%o1,%o1 + st %o1,[%o0] + retl + xnor %g0,%o2,%o0 + +LLnlop: bcc LLp1 + addxcc %o2,%o2,%o2 +LLn1: addx %o4,%o4,%o4 + subcc %o4,%g3,%o1 + bcc LLp2 + addxcc %o2,%o2,%o2 +LLn2: addx %o4,%o4,%o4 + subcc %o4,%g3,%o1 + bcc LLp3 + addxcc %o2,%o2,%o2 +LLn3: addx %o4,%o4,%o4 + subcc %o4,%g3,%o1 + bcc LLp4 + addxcc %o2,%o2,%o2 +LLn4: addx %o4,%o4,%o4 + addcc %g1,-1,%g1 + bne LLnlop + subcc %o4,%g3,%o1 + bcc LLp5 + addxcc %o2,%o2,%o2 +LLn5: add %o4,%o4,%o4 ! << 1 + tst %g2 + bne Oddn + add %o5,%o4,%o4 + st %o4,[%o0] + retl + xnor %g0,%o2,%o0 + +Oddp: xnor %g0,%o2,%o2 + ! q' in %o2. r' in %o1 + addcc %o1,%o2,%o1 + bcc LLp6 + addx %o2,0,%o2 + sub %o1,%o3,%o1 +LLp6: subcc %o1,%o3,%g0 + bcs LLp7 + subx %o2,-1,%o2 + sub %o1,%o3,%o1 +LLp7: st %o1,[%o0] + retl + mov %o2,%o0 + +Oddn: xnor %g0,%o2,%o2 + ! q' in %o2. r' in %o4 + addcc %o4,%o2,%o4 + bcc LLn6 + addx %o2,0,%o2 + sub %o4,%o3,%o4 +LLn6: subcc %o4,%o3,%g0 + bcs LLn7 + subx %o2,-1,%o2 + sub %o4,%o3,%o4 +LLn7: st %o4,[%o0] + retl + mov %o2,%o0 diff --git a/sysdeps/sparc/sparc32/sub_n.S b/sysdeps/sparc/sparc32/sub_n.S new file mode 100644 index 0000000000..b7a11958e2 --- /dev/null +++ b/sysdeps/sparc/sparc32/sub_n.S @@ -0,0 +1,311 @@ +! SPARC __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and +! store difference in a third limb vector. + +! Copyright (C) 1995, 1996 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +#define res_ptr %o0 +#define s1_ptr %o1 +#define s2_ptr %o2 +#define size %o3 + +#include "sysdep.h" + + .text + .align 4 + .global C_SYMBOL_NAME(__mpn_sub_n) +C_SYMBOL_NAME(__mpn_sub_n): + xor s2_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L1 ! branch if alignment differs + nop +! ** V1a ** + andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 + be L_v1 ! if no, branch + nop +/* Add least significant limb separately to align res_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + subcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr +L_v1: addx %g0,%g0,%o4 ! save cy in register + cmp size,2 ! if size < 2 ... + bl Lend2 ! ... branch to tail code + subcc %g0,%o4,%g0 ! restore cy + + ld [s1_ptr+0],%g4 + addcc size,-10,size + ld [s1_ptr+4],%g1 + ldd [s2_ptr+0],%g2 + blt Lfin1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop1: subxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + subxcc %g4,%g2,%o4 + ld [s1_ptr+16],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+20],%g1 + ldd [s2_ptr+16],%g2 + std %o4,[res_ptr+8] + subxcc %g4,%g2,%o4 + ld [s1_ptr+24],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+28],%g1 + ldd [s2_ptr+24],%g2 + std %o4,[res_ptr+16] + subxcc %g4,%g2,%o4 + ld [s1_ptr+32],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+36],%g1 + ldd [s2_ptr+32],%g2 + std %o4,[res_ptr+24] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop1 + subcc %g0,%o4,%g0 ! restore cy + +Lfin1: addcc size,8-2,size + blt Lend1 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1: subxcc %g4,%g2,%o4 + ld [s1_ptr+8],%g4 + subxcc %g1,%g3,%o5 + ld [s1_ptr+12],%g1 + ldd [s2_ptr+8],%g2 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope1 + subcc %g0,%o4,%g0 ! restore cy +Lend1: subxcc %g4,%g2,%o4 + subxcc %g1,%g3,%o5 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + + andcc size,1,%g0 + be Lret1 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ + ld [s1_ptr+8],%g4 + ld [s2_ptr+8],%g2 + subxcc %g4,%g2,%o4 + st %o4,[res_ptr+8] + +Lret1: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb + +L1: xor s1_ptr,res_ptr,%g1 + andcc %g1,4,%g0 + bne L2 + nop +! ** V1b ** + andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 + be L_v1b ! if no, branch + nop +/* Add least significant limb separately to align res_ptr and s1_ptr */ + ld [s2_ptr],%g4 + add s2_ptr,4,s2_ptr + ld [s1_ptr],%g2 + add s1_ptr,4,s1_ptr + add size,-1,size + subcc %g2,%g4,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr +L_v1b: addx %g0,%g0,%o4 ! save cy in register + cmp size,2 ! if size < 2 ... + bl Lend2 ! ... branch to tail code + subcc %g0,%o4,%g0 ! restore cy + + ld [s2_ptr+0],%g4 + addcc size,-10,size + ld [s2_ptr+4],%g1 + ldd [s1_ptr+0],%g2 + blt Lfin1b + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop1b: subxcc %g2,%g4,%o4 + ld [s2_ptr+8],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+12],%g1 + ldd [s1_ptr+8],%g2 + std %o4,[res_ptr+0] + subxcc %g2,%g4,%o4 + ld [s2_ptr+16],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+20],%g1 + ldd [s1_ptr+16],%g2 + std %o4,[res_ptr+8] + subxcc %g2,%g4,%o4 + ld [s2_ptr+24],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+28],%g1 + ldd [s1_ptr+24],%g2 + std %o4,[res_ptr+16] + subxcc %g2,%g4,%o4 + ld [s2_ptr+32],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+36],%g1 + ldd [s1_ptr+32],%g2 + std %o4,[res_ptr+24] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop1b + subcc %g0,%o4,%g0 ! restore cy + +Lfin1b: addcc size,8-2,size + blt Lend1b + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 2 limbs until less than 2 limbs remain */ +Loope1b:subxcc %g2,%g4,%o4 + ld [s2_ptr+8],%g4 + subxcc %g3,%g1,%o5 + ld [s2_ptr+12],%g1 + ldd [s1_ptr+8],%g2 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope1b + subcc %g0,%o4,%g0 ! restore cy +Lend1b: subxcc %g2,%g4,%o4 + subxcc %g3,%g1,%o5 + std %o4,[res_ptr+0] + addx %g0,%g0,%o4 ! save cy in register + + andcc size,1,%g0 + be Lret1b + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ + ld [s2_ptr+8],%g4 + ld [s1_ptr+8],%g2 + subxcc %g2,%g4,%o4 + st %o4,[res_ptr+8] + +Lret1b: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb + +! ** V2 ** +/* If we come here, the alignment of s1_ptr and res_ptr as well as the + alignment of s2_ptr and res_ptr differ. Since there are only two ways + things can be aligned (that we care about) we now know that the alignment + of s1_ptr and s2_ptr are the same. */ + +L2: cmp size,1 + be Ljone + nop + andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 + be L_v2 ! if no, branch + nop +/* Add least significant limb separately to align s1_ptr and s2_ptr */ + ld [s1_ptr],%g4 + add s1_ptr,4,s1_ptr + ld [s2_ptr],%g2 + add s2_ptr,4,s2_ptr + add size,-1,size + subcc %g4,%g2,%o4 + st %o4,[res_ptr] + add res_ptr,4,res_ptr + +L_v2: addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + blt Lfin2 + subcc %g0,%o4,%g0 ! restore cy +/* Add blocks of 8 limbs until less than 8 limbs remain */ +Loop2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + ldd [s1_ptr+8],%g2 + ldd [s2_ptr+8],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+8] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+12] + ldd [s1_ptr+16],%g2 + ldd [s2_ptr+16],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+16] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+20] + ldd [s1_ptr+24],%g2 + ldd [s2_ptr+24],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+24] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+28] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-8,size + add s1_ptr,32,s1_ptr + add s2_ptr,32,s2_ptr + add res_ptr,32,res_ptr + bge Loop2 + subcc %g0,%o4,%g0 ! restore cy + +Lfin2: addcc size,8-2,size + blt Lend2 + subcc %g0,%o4,%g0 ! restore cy +Loope2: ldd [s1_ptr+0],%g2 + ldd [s2_ptr+0],%o4 + subxcc %g2,%o4,%g2 + st %g2,[res_ptr+0] + subxcc %g3,%o5,%g3 + st %g3,[res_ptr+4] + addx %g0,%g0,%o4 ! save cy in register + addcc size,-2,size + add s1_ptr,8,s1_ptr + add s2_ptr,8,s2_ptr + add res_ptr,8,res_ptr + bge Loope2 + subcc %g0,%o4,%g0 ! restore cy +Lend2: andcc size,1,%g0 + be Lret2 + subcc %g0,%o4,%g0 ! restore cy +/* Add last limb */ +Ljone: ld [s1_ptr],%g4 + ld [s2_ptr],%g2 + subxcc %g4,%g2,%o4 + st %o4,[res_ptr] + +Lret2: retl + addx %g0,%g0,%o0 ! return carry-out from most sign. limb diff --git a/sysdeps/sparc/sparc32/submul_1.S b/sysdeps/sparc/sparc32/submul_1.S new file mode 100644 index 0000000000..a8ebd501a7 --- /dev/null +++ b/sysdeps/sparc/sparc32/submul_1.S @@ -0,0 +1,147 @@ +! SPARC __mpn_submul_1 -- Multiply a limb vector with a limb and subtract +! the result from a second limb vector. + +! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +! MA 02111-1307, USA. + + +! INPUT PARAMETERS +! res_ptr o0 +! s1_ptr o1 +! size o2 +! s2_limb o3 + +#include "sysdep.h" + +.text + .align 4 + .global C_SYMBOL_NAME(__mpn_submul_1) +C_SYMBOL_NAME(__mpn_submul_1): + ! Make S1_PTR and RES_PTR point at the end of their blocks + ! and put (- 4 x SIZE) in index/loop counter. + sll %o2,2,%o2 + add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval + add %o1,%o2,%o1 + sub %g0,%o2,%o2 + + cmp %o3,0xfff + bgu Large + nop + + ld [%o1+%o2],%o5 + mov 0,%o0 + b L0 + add %o4,-4,%o4 +Loop0: + subcc %o5,%g1,%g1 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g1,[%o4+%o2] +L0: wr %g0,%o3,%y + sra %o5,31,%g2 + and %o3,%g2,%g2 + andcc %g1,0,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,%o5,%g1 + mulscc %g1,0,%g1 + sra %g1,20,%g4 + sll %g1,12,%g1 + rd %y,%g3 + srl %g3,20,%g3 + or %g1,%g3,%g1 + + addcc %g1,%o0,%g1 + addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb + addcc %o2,4,%o2 ! loop counter + bne Loop0 + ld [%o4+%o2],%o5 + + subcc %o5,%g1,%g1 + addx %o0,%g0,%o0 + retl + st %g1,[%o4+%o2] + + +Large: ld [%o1+%o2],%o5 + mov 0,%o0 + sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 + b L1 + add %o4,-4,%o4 +Loop: + subcc %o5,%g3,%g3 + ld [%o1+%o2],%o5 + addx %o0,%g0,%o0 + st %g3,[%o4+%o2] +L1: wr %g0,%o5,%y + and %o5,%g4,%g2 + andcc %g0,%g0,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%g0,%g1 + rd %y,%g3 + addcc %g3,%o0,%g3 + addx %g2,%g1,%o0 + addcc %o2,4,%o2 + bne Loop + ld [%o4+%o2],%o5 + + subcc %o5,%g3,%g3 + addx %o0,%g0,%o0 + retl + st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/sparc32/udiv.S b/sysdeps/sparc/sparc32/udiv.S new file mode 100644 index 0000000000..4a7d1526b0 --- /dev/null +++ b/sysdeps/sparc/sparc32/udiv.S @@ -0,0 +1,352 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .udiv name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include +#else +#ifdef __svr4__ +#include +#else +#include +#endif +#endif + +ENTRY(.udiv) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +Lgot_result: + + retl + mov %o2, %o0 diff --git a/sysdeps/sparc/sparc32/udiv_qrnnd.S b/sysdeps/sparc/sparc32/udiv_qrnnd.S new file mode 100644 index 0000000000..5b67f874c3 --- /dev/null +++ b/sysdeps/sparc/sparc32/udiv_qrnnd.S @@ -0,0 +1,165 @@ +! SPARC __udiv_qrnnd division support, used from longlong.h. + +! Copyright (C) 1993, 1994, 1997 Free Software Foundation, Inc. + +! This file is part of the GNU MP Library. + +! The GNU MP Library is free software; you can redistribute it and/or modify +! it under the terms of the GNU Library General Public License as published by +! the Free Software Foundation; either version 2 of the License, or (at your +! option) any later version. + +! The GNU MP Library is distributed in the hope that it will be useful, but +! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +! License for more details. + +! You should have received a copy of the GNU Library General Public License +! along with the GNU MP Library; see the file COPYING.LIB. If not, write to +! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + +! Added PIC support - May/96, Miguel de Icaza + +! INPUT PARAMETERS +! rem_ptr i0 +! n1 i1 +! n0 i2 +! d i3 + +#include "sysdep.h" +#undef ret /* Kludge for glibc */ + + .text + .align 8 + + .type two_to_32,@object +two_to_32: + .double 0r4294967296 + .size two_to_32,8 + + .type two_to_31,@object +two_to_31: + .double 0r2147483648 + .size two_to_31,8 + + .align 4 + .global __udiv_qrnnd + .type __udiv_qrnnd,@function +ENTRY(__udiv_qrnnd) + !#PROLOGUE# 0 + save %sp,-104,%sp + !#PROLOGUE# 1 + st %i1,[%fp-8] + ld [%fp-8],%f10 +#ifdef PIC +.Lbase: call 1f + fitod %f10,%f4 +1: ldd [%o7-(.Lbase-two_to_32)],%f8 +#else + sethi %hi(two_to_32),%o7 + fitod %f10,%f4 + ldd [%o7+%lo(two_to_32)],%f8 +#endif + cmp %i1,0 + bge L248 + mov %i0,%i5 + faddd %f4,%f8,%f4 +.L248: + st %i2,[%fp-8] + ld [%fp-8],%f10 + fmuld %f4,%f8,%f6 + cmp %i2,0 + bge L249 + fitod %f10,%f2 + faddd %f2,%f8,%f2 +.L249: + st %i3,[%fp-8] + faddd %f6,%f2,%f2 + ld [%fp-8],%f10 + cmp %i3,0 + bge L250 + fitod %f10,%f4 + faddd %f4,%f8,%f4 +.L250: + fdivd %f2,%f4,%f2 +#ifdef PIC + ldd [%o7-(.Lbase-two_to_31)],%f4 +#else + sethi %hi(two_to_31),%o7 + ldd [%o7+%lo(two_to_31)],%f4 +#endif + fcmped %f2,%f4 + nop + fbge,a L251 + fsubd %f2,%f4,%f2 + fdtoi %f2,%f2 + st %f2,[%fp-8] + b L252 + ld [%fp-8],%i4 +.L251: + fdtoi %f2,%f2 + st %f2,[%fp-8] + ld [%fp-8],%i4 + sethi %hi(-2147483648),%g2 + xor %i4,%g2,%i4 +.L252: + wr %g0,%i4,%y + sra %i3,31,%g2 + and %i4,%g2,%g2 + andcc %g0,0,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,%i3,%g1 + mulscc %g1,0,%g1 + add %g1,%g2,%i0 + rd %y,%g3 + subcc %i2,%g3,%o7 + subxcc %i1,%i0,%g0 + be L253 + cmp %o7,%i3 + + add %i4,-1,%i0 + add %o7,%i3,%o7 + st %o7,[%i5] + ret + restore +.L253: + blu L246 + mov %i4,%i0 + add %i4,1,%i0 + sub %o7,%i3,%o7 +.L246: + st %o7,[%i5] + ret + restore + + .size __udiv_qrnnd, .-__udiv_qrnnd diff --git a/sysdeps/sparc/sparc32/umul.S b/sysdeps/sparc/sparc32/umul.S new file mode 100644 index 0000000000..7a26c295cb --- /dev/null +++ b/sysdeps/sparc/sparc32/umul.S @@ -0,0 +1,153 @@ +/* + * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the + * upper 32 bits of the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. Short + * multiplies require 25 instruction cycles, and long ones require + * 45 instruction cycles. + * + * On return, overflow has occurred (%o1 is not zero) if and only if + * the Z condition code is clear, allowing, e.g., the following: + * + * call .umul + * nop + * bnz overflow (or tnz) + */ + +#include "DEFS.h" +FUNC(.umul) + or %o0, %o1, %o4 + mov %o0, %y ! multiplier -> Y + andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + + /* + * Normally, with the shift-and-add approach, if both numbers are + * positive you get the correct result. With 32-bit two's-complement + * numbers, -x is represented as + * + * x 32 + * ( 2 - ------ ) mod 2 * 2 + * 32 + * 2 + * + * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s, + * we can treat this as if the radix point were just to the left + * of the sign bit (multiply by 2^32), and get + * + * -x = (2 - x) mod 2 + * + * Then, ignoring the `mod 2's for convenience: + * + * x * y = xy + * -x * y = 2y - xy + * x * -y = 2x - xy + * -x * -y = 4 - 2x - 2y + xy + * + * For signed multiplies, we subtract (x << 32) from the partial + * product to fix this problem for negative multipliers (see mul.s). + * Because of the way the shift into the partial product is calculated + * (N xor V), this term is automatically removed for the multiplicand, + * so we don't have to adjust. + * + * But for unsigned multiplies, the high order bit wasn't a sign bit, + * and the correction is wrong. So for unsigned multiplies where the + * high order bit is one, we end up with xy - (y << 32). To fix it + * we add y << 32. + */ +#if 0 + tst %o1 + bl,a 1f ! if %o1 < 0 (high order bit = 1), + add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half) +1: rd %y, %o0 ! get lower half of product + retl + addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0 +#else + /* Faster code from tege@sics.se. */ + sra %o1, 31, %o2 ! make mask from sign bit + and %o0, %o2, %o2 ! %o2 = 0 or %o0, depending on sign of %o1 + rd %y, %o0 ! get lower half of product + retl + addcc %o4, %o2, %o1 ! add compensation and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above), + * and overflow is impossible (the answer is at most 24 bits long). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the result; %y has + * the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * -----result----- + * + * The 12 bits of %o4 left of the `result' area are all zero; + * in fact, all top 20 bits of %o4 are zero. + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20 + or %o5, %o0, %o0 + retl + addcc %g0, %g0, %o1 ! %o1 = zero, and set Z diff --git a/sysdeps/sparc/sparc32/urem.S b/sysdeps/sparc/sparc32/urem.S new file mode 100644 index 0000000000..e72c33dd63 --- /dev/null +++ b/sysdeps/sparc/sparc32/urem.S @@ -0,0 +1,352 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .urem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "sysdep.h" +#ifdef __linux__ +#include +#else +#ifdef __svr4__ +#include +#else +#include +#endif +#endif + +ENTRY(.urem) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + + +Lgot_result: + + retl + mov %o3, %o0 diff --git a/sysdeps/sparc/sparc64/Implies b/sysdeps/sparc/sparc64/Implies new file mode 100644 index 0000000000..a8cae95f9d --- /dev/null +++ b/sysdeps/sparc/sparc64/Implies @@ -0,0 +1 @@ +wordsize-64 diff --git a/sysdeps/sparc/sparc64/Makefile b/sysdeps/sparc/sparc64/Makefile new file mode 100644 index 0000000000..656274d75c --- /dev/null +++ b/sysdeps/sparc/sparc64/Makefile @@ -0,0 +1,11 @@ +# The Sparc`long double' is a distinct type we support. +#long-double-fcts = yes + +# But the support for ieee quads is so bad we just skip the whole thing. +# XXX This is only a temporary hack. +omit-long-double-fcts = yes + +# But we do need these few functions to even link stdio proggies +ifeq ($(subdir),math) +sysdep_routines += s_isinfl s_isnanl s_finitel s_signbitl +endif diff --git a/sysdeps/sparc/sparc64/add_n.S b/sysdeps/sparc/sparc64/add_n.S new file mode 100644 index 0000000000..68bb008a7d --- /dev/null +++ b/sysdeps/sparc/sparc64/add_n.S @@ -0,0 +1,58 @@ +/* SPARC v9 __mpn_add_n -- Add two limb vectors of the same length > 0 and + store sum in a third limb vector. + + Copyright (C) 1995, 1996, 1997 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + + +#include + + +/* INPUT PARAMETERS + res_ptr %o0 + s1_ptr %o1 + s2_ptr %o2 + size %o3 */ + + +ENTRY(__mpn_add_n) + + sub %g0,%o3,%g3 + sllx %o3,3,%g1 + add %o1,%g1,%o1 ! make s1_ptr point at end + add %o2,%g1,%o2 ! make s2_ptr point at end + add %o0,%g1,%o0 ! make res_ptr point at end + mov 0,%o4 ! clear carry variable + sllx %g3,3,%o5 ! compute initial address index + +1: ldx [%o2+%o5],%g1 ! load s2 limb + add %g3,1,%g3 ! increment loop count + ldx [%o1+%o5],%g2 ! load s1 limb + addcc %g1,%o4,%g1 ! add s2 limb and carry variable + movcc %xcc,0,%o4 ! if carry-out, o4 was 1; clear it + addcc %g1,%g2,%g1 ! add s1 limb to sum + stx %g1,[%o0+%o5] ! store result + add %o5,8,%o5 ! increment address index + brnz,pt %g3,1b + movcs %xcc,1,%o4 ! if s1 add gave carry, record it + + retl + mov %o4,%o0 + +END(__mpn_add_n) diff --git a/sysdeps/sparc/sparc64/addmul_1.S b/sysdeps/sparc/sparc64/addmul_1.S new file mode 100644 index 0000000000..260481230f --- /dev/null +++ b/sysdeps/sparc/sparc64/addmul_1.S @@ -0,0 +1,88 @@ +/* SPARC v9 __mpn_addmul_1 -- Multiply a limb vector with a single limb and + add the product to a second limb vector. + + Copyright (C) 1996 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + +#include + + +/* INPUT PARAMETERS + res_ptr o0 + s1_ptr o1 + size o2 + s2_limb o3 */ + + +ENTRY(__mpn_addmul_1) + !#PROLOGUE# 0 + save %sp,-128,%sp + !#PROLOGUE# 1 + + sub %g0,%i2,%o7 + sllx %o7,3,%g5 + sub %i1,%g5,%o3 + sub %i0,%g5,%o4 + mov 0,%o0 ! zero cy_limb + + srl %i3,0,%o1 ! extract low 32 bits of s2_limb + srlx %i3,32,%i3 ! extract high 32 bits of s2_limb + mov 1,%o2 + sllx %o2,32,%o2 ! o2 = 0x100000000 + + ! hi ! + ! mid-1 ! + ! mid-2 ! + ! lo ! +1: + sllx %o7,3,%g1 + ldx [%o3+%g1],%g5 + srl %g5,0,%i0 ! zero hi bits + srlx %g5,32,%g5 + mulx %o1,%i0,%i4 ! lo product + mulx %i3,%i0,%i1 ! mid-1 product + mulx %o1,%g5,%l2 ! mid-2 product + mulx %i3,%g5,%i5 ! hi product + srlx %i4,32,%i0 ! extract high 32 bits of lo product... + add %i1,%i0,%i1 ! ...and add it to the mid-1 product + addcc %i1,%l2,%i1 ! add mid products + mov 0,%l0 ! we need the carry from that add... + movcs %xcc,%o2,%l0 ! ...compute it and... + add %i5,%l0,%i5 ! ...add to bit 32 of the hi product + sllx %i1,32,%i0 ! align low bits of mid product + srl %i4,0,%g5 ! zero high 32 bits of lo product + add %i0,%g5,%i0 ! combine into low 64 bits of result + srlx %i1,32,%i1 ! extract high bits of mid product... + add %i5,%i1,%i1 ! ...and add them to the high result + addcc %i0,%o0,%i0 ! add cy_limb to low 64 bits of result + mov 0,%g5 + movcs %xcc,1,%g5 + add %o7,1,%o7 + ldx [%o4+%g1],%l1 + addcc %l1,%i0,%i0 + movcs %xcc,1,%g5 + stx %i0,[%o4+%g1] + brnz %o7,1b + add %i1,%g5,%o0 ! compute new cy_limb + + mov %o0,%i0 + jmpl %i7+8, %g0 + restore + +END(__mpn_addmul_1) diff --git a/sysdeps/sparc/sparc64/bits/endian.h b/sysdeps/sparc/sparc64/bits/endian.h new file mode 100644 index 0000000000..a2ab07249a --- /dev/null +++ b/sysdeps/sparc/sparc64/bits/endian.h @@ -0,0 +1,8 @@ +/* Sparc is big-endian, but v9 supports endian conversion on loads/stores + and GCC supports such a mode. Be prepared. */ + +#ifdef __LITTLE_ENDIAN__ +#define __BYTE_ORDER __LITTLE_ENDIAN +#else +#define __BYTE_ORDER __BIG_ENDIAN +#endif diff --git a/sysdeps/sparc/sparc64/dl-machine.h b/sysdeps/sparc/sparc64/dl-machine.h new file mode 100644 index 0000000000..eb46ba92be --- /dev/null +++ b/sysdeps/sparc/sparc64/dl-machine.h @@ -0,0 +1,353 @@ +/* Machine-dependent ELF dynamic relocation inline functions. Sparc64 version. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If + not, write to the Free Software Foundation, Inc., + 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + +#define ELF_MACHINE_NAME "sparc64" + +#include +#include +#include +#include +#include + + +/* Translate a processor-specific dynamic tag to the index into l_info. */ +#define DT_SPARC(x) (DT_SPARC_##x - DT_LOPROC + DT_NUM) + +/* Return nonzero iff E_MACHINE is compatible with the running host. */ +static inline int +elf_machine_matches_host (Elf64_Half e_machine) +{ + return e_machine == EM_SPARC64; +} + +/* Return the link-time address of _DYNAMIC. Conveniently, this is the + first element of the GOT. This must be inlined in a function which + uses global data. */ +static inline Elf64_Addr +elf_machine_dynamic (void) +{ + register Elf64_Addr elf_pic_register __asm__("%l7"); + + return *(Elf64_Addr *)elf_pic_register; +} + + +/* Return the run-time load address of the shared object. */ +static inline Elf64_Addr +elf_machine_load_address (void) +{ + register Elf64_Addr elf_pic_register __asm__("%l7"); + Elf64_Addr pc, la; + + /* Utilize the fact that a local .got entry will be partially + initialized at startup awaiting its RELATIVE fixup. */ + + __asm("sethi %%hi(.Load_address), %1\n" + ".Load_address:\n\t" + "rd %%pc, %0\n\t" + "or %1, %%lo(.Load_address), %1\n\t" + : "=r"(pc), "=r"(la)); + + return pc - *(Elf64_Addr *)(elf_pic_register + la); +} + +#ifdef RESOLVE +/* Perform the relocation specified by RELOC and SYM (which is fully resolved). + MAP is the object containing the reloc. */ + +static inline void +elf_machine_rela (struct link_map *map, const Elf64_Rela *reloc, + const Elf64_Sym *sym, const struct r_found_version *version, + Elf64_Addr *const reloc_addr) +{ +#ifndef RTLD_BOOTSTRAP + /* This is defined in rtld.c, but nowhere in the static libc.a; make the + reference weak so static programs can still link. This declaration + cannot be done when compiling rtld.c (i.e. #ifdef RTLD_BOOTSTRAP) + because rtld.c contains the common defn for _dl_rtld_map, which is + incompatible with a weak decl in the same file. */ + weak_extern (_dl_rtld_map); +#endif + + if (ELF64_R_TYPE (reloc->r_info) == R_SPARC_RELATIVE) + { +#ifndef RTLD_BOOTSTRAP + if (map != &_dl_rtld_map) /* Already done in rtld itself. */ +#endif + *reloc_addr = map->l_addr + reloc->r_addend; + } + else + { + const Elf64_Sym *const refsym = sym; + Elf64_Addr value; + if (sym->st_shndx != SHN_UNDEF && + ELF64_ST_BIND (sym->st_info) == STB_LOCAL) + value = map->l_addr; + else + { + value = RESOLVE (&sym, version, ELF64_R_TYPE (reloc->r_info)); + if (sym) + value += sym->st_value; + } + value += reloc->r_addend; /* Assume copy relocs have zero addend. */ + + switch (ELF64_R_TYPE (reloc->r_info)) + { + case R_SPARC_COPY: + if (sym->st_size > refsym->st_size + || (_dl_verbose && sym->st_size < refsym->st_size)) + { + extern char **_dl_argv; + const char *strtab; + + strtab = ((void *) map->l_addr + + map->l_info[DT_STRTAB]->d_un.d_ptr); + _dl_sysdep_error (_dl_argv[0] ?: "", + ": Symbol `", strtab + refsym->st_name, + "' has different size in shared object, " + "consider re-linking\n", NULL); + } + memcpy (reloc_addr, (void *) value, MIN (sym->st_size, + refsym->st_size)); + break; + + case R_SPARC_64: + case R_SPARC_GLOB_DAT: + *reloc_addr = value; + break; + case R_SPARC_8: + *(char *) reloc_addr = value; + break; + case R_SPARC_16: + *(short *) reloc_addr = value; + break; + case R_SPARC_DISP8: + *(char *) reloc_addr = (value - (Elf64_Addr) reloc_addr); + break; + case R_SPARC_DISP16: + *(short *) reloc_addr = (value - (Elf64_Addr) reloc_addr); + break; + case R_SPARC_DISP32: + *(unsigned int *)reloc_addr = (value - (Elf64_Addr) reloc_addr); + break; + case R_SPARC_LO10: + *(unsigned *)reloc_addr = (*(unsigned *)reloc_addr & ~0x3ff) + | (value & 0x3ff); + break; + case R_SPARC_WDISP30: + *(unsigned *)reloc_addr = ((*(unsigned *)reloc_addr & 0xc0000000) + | ((value - (Elf64_Addr) reloc_addr) >> 2)); + break; + case R_SPARC_HI22: + *(unsigned *)reloc_addr = (*(unsigned *)reloc_addr & 0xffc00000) + | (value >> 10); + break; + + case R_SPARC_JMP_SLOT: + { + Elf64_Dyn *pltfmt = map->l_info[DT_SPARC(PLTFMT)]; + switch (pltfmt ? pltfmt->d_un.d_val : 0) + { + case 1: /* .got.plt with absolute addresses */ + *reloc_addr = value; + break; + case 2: /* .got.plt with got-relative addresses */ + *reloc_addr = value - (map->l_info[DT_PLTGOT]->d_un.d_ptr + + map->l_addr); + break; + default: + assert (! "unexpected .plt format type"); + } + } + break; + + case R_SPARC_NONE: /* Alright, Wilbur. */ + break; + default: + assert (! "unexpected dynamic reloc type"); + break; + } + } +} + +static inline void +elf_machine_lazy_rel (struct link_map *map, const Elf64_Rela *reloc) +{ + switch (ELF64_R_TYPE (reloc->r_info)) + { + case R_SPARC_NONE: + break; + case R_SPARC_JMP_SLOT: + break; + default: + assert (! "unexpected PLT reloc type"); + break; + } +} + +#endif /* RESOLVE */ + +/* Nonzero iff TYPE should not be allowed to resolve to one of + the main executable's symbols, as for a COPY reloc. */ +#define elf_machine_lookup_noexec_p(type) ((type) == R_SPARC_COPY) + +/* Nonzero iff TYPE describes relocation of a PLT entry, so + PLT entries should not be allowed to define the value. */ +#define elf_machine_lookup_noplt_p(type) ((type) == R_SPARC_JMP_SLOT) + +/* A reloc type used for ld.so cmdline arg lookups to reject PLT entries. */ +#define ELF_MACHINE_RELOC_NOPLT R_SPARC_JMP_SLOT + +/* The SPARC never uses Elf64_Rel relocations. */ +#define ELF_MACHINE_NO_REL 1 + +/* The SPARC overlaps DT_RELA and DT_PLTREL. */ +#define ELF_MACHINE_PLTREL_OVERLAP 1 + +/* Set up the loaded object described by L so its unrelocated PLT + entries will jump to the on-demand fixup code in dl-runtime.c. */ + +static inline int +elf_machine_runtime_setup (struct link_map *l, int lazy, int profile) +{ + Elf64_Addr *got; + extern void _dl_runtime_resolve (void); + + if (l->l_info[DT_JMPREL] && lazy) + { + got = (Elf64_Addr *) (l->l_addr + l->l_info[DT_PLTGOT]->d_un.d_ptr); + got[1] = (Elf64_Addr) l; /* Identify this shared object. */ + /* This function will get called to fix up the GOT entry indicated by + the offset on the stack, and then jump to the resolved address. */ + got[2] = (Elf64_Addr) &_dl_runtime_resolve; + } + + return lazy; +} + +/* This code is used in dl-runtime.c to call the `fixup' function + and then redirect to the address it returns. */ +#define ELF_MACHINE_RUNTIME_TRAMPOLINE asm ("\ + .globl _dl_runtime_resolve + .type _dl_runtime_resolve, @function +_dl_runtime_resolve: + save %sp, -160, %sp + mov %g5, %o0 + call fixup + mov %g6, %o1 + jmp %o0 + restore + .size _dl_runtime_resolve, .-_dl_runtime_resolve +"); + +/* The PLT uses Elf64_Rela relocs. */ +#define elf_machine_relplt elf_machine_rela + + +/* Initial entry point code for the dynamic linker. + The C function `_dl_start' is the real entry point; + its return value is the user program's entry point. */ + +#define __S1(x) #x +#define __S(x) __S1(x) + +#define RTLD_START __asm__ ( "\ + .global _start + .type _start, @function +_start: + /* Make room for functions to drop their arguments on the stack. */ + sub %sp, 6*8, %sp + /* Pass pointer to argument block to _dl_start. */ + call _dl_start + add %sp," __S(STACK_BIAS) "+22*8,%o0 + /* FALLTHRU */ + .size _start, .-_start + + .global _dl_start_user + .type _dl_start_user, @function +_dl_start_user: + /* Load the GOT register. */ +1: call 11f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)),%l7 +11: or %l7,%lo(_GLOBAL_OFFSET_TABLE_-(1b-.)),%l7 + add %l7,%o7,%l7 + /* Save the user entry point address in %l0. */ + mov %o0,%l0 + /* See if we were run as a command with the executable file name as an + extra leading argument. If so, we must shift things around since we + must keep the stack doubleword aligned. */ + sethi %hi(_dl_skip_args), %g2 + or %g2, %lo(_dl_skip_args), %g2 + ldx [%l7+%g2], %i0 + ld [%i0], %i0 + brz,pt %i0, 2f + nop + /* Find out how far to shift. */ + ldx [%sp+" __S(STACK_BIAS) "+22*8], %i1 + sub %i1, %i0, %i1 + sllx %i0, 3, %i2 + stx %i1, [%sp+" __S(STACK_BIAS) "+22*8] + add %sp, " __S(STACK_BIAS) "+23*8, %i1 + add %i1, %i2, %i2 + /* Copy down argv. */ +12: ldx [%i2], %i3 + add %i2, 8, %i2 + stx %i3, [%i1] + brnz,pt %i3, 12b + add %i1, 8, %i1 + /* Copy down envp. */ +13: ldx [%i2], %i3 + add %i2, 8, %i2 + stx %i3, [%i1] + brnz,pt %i3, 13b + add %i1, 8, %i1 + /* Copy down auxiliary table. */ +14: ldx [%i2], %i3 + ldx [%i2+8], %i4 + add %i2, 16, %i2 + stx %i3, [%i1] + stx %i4, [%i1+8] + brnz,pt %i3, 13b + add %i1, 16, %i1 + /* Load _dl_default_scope[2] to pass to _dl_init_next. */ +2: sethi %hi(_dl_default_scope), %g2 + or %g2, %lo(_dl_default_scope), %g2 + ldx [%l7+%g2], %g2 + ldx [%g2+2*8], %l1 + /* Call _dl_init_next to return the address of an initializer to run. */ +3: call _dl_init_next + mov %l1, %o0 + brz,pn %o0, 4f + nop + jmpl %o0, %o7 + nop + ba,a 3b + /* Clear the startup flag. */ +4: sethi %hi(_dl_starting_up), %g2 + or %g2, %lo(_dl_starting_up), %g2 + ldx [%l7+%g2], %g2 + st %g0, [%g2] + /* Pass our finalizer function to the user in %g1. */ + sethi %hi(_dl_fini), %g1 + or %g1, %lo(_dl_fini), %g1 + ldx [%l7+%g1], %g1 + /* Jump to the user's entry point & undo the allocation of the xtra regs. */ + jmp %l0 + add %sp, 6*8, %sp + .size _dl_start_user, .-_dl_start_user"); diff --git a/sysdeps/sparc/sparc64/elf/Dist b/sysdeps/sparc/sparc64/elf/Dist new file mode 100644 index 0000000000..d9338c8fd6 --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/Dist @@ -0,0 +1,4 @@ +crtbegin.S +crtbeginS.S +crtend.S +crtendS.S diff --git a/sysdeps/sparc/sparc64/elf/Makefile b/sysdeps/sparc/sparc64/elf/Makefile new file mode 100644 index 0000000000..1b38355b38 --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/Makefile @@ -0,0 +1,10 @@ +# The assembler on SPARC needs the -fPIC flag even when it's assembler code. +ASFLAGS-.so = -fPIC + +ifeq ($(subdir), csu) +extra-objs += crtbegin.o crtend.o crtbeginS.o crtendS.o +install-lib += crtbegin.o crtend.o crtbeginS.o crtendS.o + +CPPFLAGS-crtbeginS.S = -fPIC -DPIC +CPPFLAGS-crtendS.S = -fPIC -DPIC +endif diff --git a/sysdeps/sparc/sparc64/elf/crtbegin.S b/sysdeps/sparc/sparc64/elf/crtbegin.S new file mode 100644 index 0000000000..318e7a6a3f --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/crtbegin.S @@ -0,0 +1,69 @@ +/* Destructor cleanup code for elf64-sparc + Copyright (C) 1997 Free Software Foundation, Inc. + Contributed by Richard Henderson , 1997. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + + +.section ".ctors",#alloc,#write + + .align 8 +__CTOR_LIST__: + .xword -1 + +.section ".dtors",#alloc,#write + + .align 8 +__DTOR_LIST__: + .xword -1 + +.section ".fini",#alloc,#execinstr + + call __do_global_dtors_aux + nop + +.text + + .align 4 + .type __do_global_dtors_aux,#function +__do_global_dtors_aux: + save %sp,-160,%sp + +#ifdef PIC +1: call 11f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %l7 +11: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %l7 + add %l7, %o7, %l7 + sethi %hi(__DTOR_LIST__), %l0 + or %l0, %lo(__DTOR_LIST__), %l0 + ldx [%l7+%l0], %l0 +#else + sethi %hi(__DTOR_LIST__), %l0 + or %l0, %lo(__DTOR_LIST__), %l0 + add %l0, %g4, %l0 +#endif + + ba 3f + ldx [%l0+8], %l1 +2: jmpl %l1, %o7 + ldx [%l0+8], %l1 +3: brnz,pt %l1, 2b + add %l0, 8, %l0 + + ret + restore + + .size __do_global_dtors_aux,.-__do_global_dtors_aux diff --git a/sysdeps/sparc/sparc64/elf/crtbeginS.S b/sysdeps/sparc/sparc64/elf/crtbeginS.S new file mode 100644 index 0000000000..7db4bc590e --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/crtbeginS.S @@ -0,0 +1 @@ +#include "crtbegin.S" diff --git a/sysdeps/sparc/sparc64/elf/crtend.S b/sysdeps/sparc/sparc64/elf/crtend.S new file mode 100644 index 0000000000..7a5c067b07 --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/crtend.S @@ -0,0 +1,70 @@ +/* Constructor startup code for elf64-sparc + Copyright (C) 1997 Free Software Foundation, Inc. + Contributed by Richard Henderson , 1997. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + + +.section ".ctors",#alloc,#write + + .align 8 +__CTOR_END__: + .xword 0 + +.section ".dtors",#alloc,#write + + .align 8 +__DTOR_END__: + .xword 0 + +.section ".init",#alloc,#execinstr + + call __do_global_ctors_aux + nop + +.text + + .align 4 + .type __do_global_ctors_aux,#function +__do_global_ctors_aux: + save %sp,-160,%sp + +#ifdef PIC +1: call 11f + sethi %hi(_GLOBAL_OFFSET_TABLE_-(1b-.)), %l7 +11: or %l7, %lo(_GLOBAL_OFFSET_TABLE_-(1b-.)), %l7 + add %l7, %o7, %l7 + sethi %hi(__CTOR_END__), %l0 + or %l0, %lo(__CTOR_END__), %l0 + ldx [%l7+%l0], %l0 +#else + sethi %hi(__CTOR_END__), %l0 + or %l0, %lo(__CTOR_END__), %l0 + add %l0, %g4, %l0 +#endif + + ba 3f + ldx [%l0+8], %l1 +2: jmpl %l1, %o7 + ldx [%l0+8], %l1 +3: addcc %l1, 1, %g0 + bnz,pt %xcc, 2b + add %l0, 8, %l0 + + ret + restore + + .size __do_global_ctors_aux,.-__do_global_ctors_aux diff --git a/sysdeps/sparc/sparc64/elf/crtendS.S b/sysdeps/sparc/sparc64/elf/crtendS.S new file mode 100644 index 0000000000..56532f567d --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/crtendS.S @@ -0,0 +1 @@ +#include "crtend.S" diff --git a/sysdeps/sparc/sparc64/elf/start.S b/sysdeps/sparc/sparc64/elf/start.S new file mode 100644 index 0000000000..73f42367ee --- /dev/null +++ b/sysdeps/sparc/sparc64/elf/start.S @@ -0,0 +1,93 @@ +/* Startup code for elf64-sparc + Copyright (C) 1997 Free Software Foundation, Inc. + Contributed by Richard Henderson , 1997. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include + + + .section ".text" + .align 4 + .global _start + .type _start,#function +_start: + + /* First order of business is to load %g4 with our base address. */ + sethi %uhi(_start), %g4 + or %g4, %ulo(_start), %g4 + sllx %g4, 32, %g4 + + /* Terminate the stack frame, and reserve space for functions to + drop their arguments. */ + mov %g0, %fp + sub %sp, 6*8, %sp + + /* Save %g1. When starting a binary via the dynamic linker, %g1 + contains the address of the shared library termination function, + which we will register below with atexit() to be called by exit(). + If we are statically linked, this will be NULL. */ + + /* Do essential libc initialization (sp points to argc, argv, and envp) */ + call __libc_init_first + mov %g1, %l0 + + /* Now that we have the proper stack frame, register library termination + function, if there is any: */ + + brz,pn %l0, 1f + nop + call atexit + mov %l0, %o0 +1: + + /* Extract the arguments and environment as encoded on the stack. The + argument info starts after one register window (16 words) past the SP, + plus the bias we added, plus the magic v9 STACK_BIAS. */ + ldx [%sp+STACK_BIAS+22*8], %o0 + add %sp, STACK_BIAS+23*8, %o1 + sllx %o0, 3, %o2 + add %o2, %o1, %o2 + sethi %hi(__environ), %g2 + add %o2, 8, %o2 + add %g2, %g4, %g2 + stx %o2, [%g2+%lo(__environ)] + + mov %o0, %l0 /* tuck them away */ + mov %o1, %l1 + + /* Call _init, the entry point to our own .init section. */ + call _init + mov %o2, %l2 + + /* Register our .fini section with atexit. */ + sethi %hi(_fini), %o0 + add %o0, %g4, %o0 + call atexit + add %o0, %lo(_fini), %o0 + + /* Call the user's main and exit with its return value. */ + mov %l0, %o0 + mov %l1, %o1 + call main + mov %l2, %o2 + call exit + nop + + /* Die very horribly if exit returns. */ + illtrap 0 + + .size _start,.-_start diff --git a/sysdeps/sparc/sparc64/fpu/bits/fenv.h b/sysdeps/sparc/sparc64/fpu/bits/fenv.h new file mode 100644 index 0000000000..802c58bdce --- /dev/null +++ b/sysdeps/sparc/sparc64/fpu/bits/fenv.h @@ -0,0 +1,76 @@ +/* Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _FENV_H +#error "Never use directly; include instead." +#endif + + +/* Define bits representing the exception. We use the bit positions + of the appropriate accrued exception bits from the FSR. */ +enum + { + FE_INVALID = (1 << 9), +#define FE_INVALID FE_INVALID + FE_OVERFLOW = (1 << 8), +#define FE_OVERFLOW FE_OVERFLOW + FE_UNDERFLOW = (1 << 7), +#define FE_UNDERFLOW FE_UNDERFLOW + FE_DIVBYZERO = (1 << 6), +#define FE_DIVBYZERO FE_DIVBYZERO + FE_INEXACT = (1 << 5) +#define FE_INEXACT FE_INEXACT + }; + +#define FE_ALL_EXCEPT \ + (FE_INEXACT | FE_DIVBYZERO | FE_UNDERFLOW | FE_OVERFLOW | FE_INVALID) + +/* The Sparc FPU supports all of the four defined rounding modes. We + use again the bit positions in the FPU control word as the values + for the appropriate macros. */ +enum + { + FE_TONEAREST = (0U << 30), +#define FE_TONEAREST FE_TONEAREST + FE_TOWARDSZERO = (1U << 30), +#define FE_TOWARDSZERO FE_TOWARDSZERO + FE_UPWARD = (2U << 30), +#define FE_UPWARD FE_UPWARD + FE_DOWNWARD = (3U << 30) +#define FE_DOWNWARD FE_DOWNWARD + }; + +#define __FE_ROUND_MASK (3U << 30) + +/* Type representing exception flags. */ +typedef unsigned long fexcept_t; + +/* Type representing floating-point environment. */ +typedef unsigned long fenv_t; + +/* If the default argument is used we use this value. */ +#define FE_DFL_ENV ((fenv_t *) -1) + +#ifdef __USE_GNU +/* Floating-point environment where none of the exception is masked. */ +# define FE_NOMASK_ENV ((fenv_t *) -2) +#endif + +/* For internal use only: access the fp state register. */ +#define __fenv_stfsr(X) __asm__("stx %%fsr,%0" : "=m"(X)) +#define __fenv_ldfsr(X) __asm__ __volatile__("ldx %0,%%fsr" : : "m"(X)) diff --git a/sysdeps/sparc/sparc64/fpu/fpu_control.h b/sysdeps/sparc/sparc64/fpu/fpu_control.h new file mode 100644 index 0000000000..fd8abb19f1 --- /dev/null +++ b/sysdeps/sparc/sparc64/fpu/fpu_control.h @@ -0,0 +1,69 @@ +/* FPU control word bits. SPARC v9 version. + Copyright (C) 1997 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Miguel de Icaza + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public License as + published by the Free Software Foundation; either version 2 of the + License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with the GNU C Library; see the file COPYING.LIB. If not, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#ifndef _FPU_CONTROL_H +#define _FPU_CONTROL_H 1 + + +#include + +/* precision control */ +#define _FPU_EXTENDED 0x00000000 /* RECOMMENDED */ +#define _FPU_DOUBLE 0x20000000 +#define _FPU_80BIT 0x30000000 +#define _FPU_SINGLE 0x10000000 /* DO NOT USE */ + +/* rounding control / Sparc */ +#define _FPU_RC_DOWN 0xc0000000 +#define _FPU_RC_UP 0x80000000 +#define _FPU_RC_ZERO 0x40000000 +#define _FPU_RC_NEAREST 0x0 /* RECOMMENDED */ + +#define _FPU_RESERVED 0x30300000 /* Reserved bits in cw */ + + +/* Now two recommended cw */ + +/* Linux default: + - extended precision + - rounding to nearest + - no exceptions */ +#define _FPU_DEFAULT 0x0 + +/* IEEE: same as above */ +#define _FPU_IEEE 0x0 + +/* Type of the control word. */ +typedef unsigned long fpu_control_t; + +#define _FPU_GETCW(cw) __asm__ ("stx %%fsr,%0" : "=m" (*&cw)) +#define _FPU_SETCW(cw) __asm__ ("ldx %0,%%fsr" : : "m" (*&cw)) + +/* Default control word set at startup. */ +extern fpu_control_t __fpu_control; + +__BEGIN_DECLS + +/* Called at startup. It can be used to manipulate fpu control register. */ +extern void __setfpucw __P ((fpu_control_t)); + +__END_DECLS + +#endif /* fpu_control.h */ diff --git a/sysdeps/sparc/sparc64/gmp-mparam.h b/sysdeps/sparc/sparc64/gmp-mparam.h new file mode 100644 index 0000000000..a3c66974de --- /dev/null +++ b/sysdeps/sparc/sparc64/gmp-mparam.h @@ -0,0 +1,27 @@ +/* gmp-mparam.h -- Compiler/machine parameter header file. + +Copyright (C) 1991, 1993, 1994 Free Software Foundation, Inc. + +This file is part of the GNU MP Library. + +The GNU MP Library is free software; you can redistribute it and/or modify +it under the terms of the GNU Library General Public License as published by +the Free Software Foundation; either version 2 of the License, or (at your +option) any later version. + +The GNU MP Library is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public +License for more details. + +You should have received a copy of the GNU Library General Public License +along with the GNU MP Library; see the file COPYING.LIB. If not, write to +the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, +MA 02111-1307, USA. */ + +#define BITS_PER_MP_LIMB 64 +#define BYTES_PER_MP_LIMB 8 +#define BITS_PER_LONGINT 64 +#define BITS_PER_INT 32 +#define BITS_PER_SHORTINT 16 +#define BITS_PER_CHAR 8 diff --git a/sysdeps/sparc/sparc64/lshift.S b/sysdeps/sparc/sparc64/lshift.S new file mode 100644 index 0000000000..1678991529 --- /dev/null +++ b/sysdeps/sparc/sparc64/lshift.S @@ -0,0 +1,96 @@ +/* SPARC v9 __mpn_lshift -- + + Copyright (C) 1996 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + +#include + +/* INPUT PARAMETERS + res_ptr %i0 + src_ptr %i1 + size %i2 + cnt %i3 */ + +ENTRY(__mpn_lshift) + save %sp, -128, %sp + + sllx %i2,3,%g1 + add %i1,%g1,%i1 ! make %i1 point at end of src + ldx [%i1-8],%g2 ! load first limb + sub %g0,%i3,%i5 ! negate shift count + add %i0,%g1,%i0 ! make %i0 point at end of res + add %i2,-1,%i2 + and %i2,4-1,%l4 ! number of limbs in first loop + srlx %g2,%i5,%g1 ! compute function result + brz,pn %l4,.L0 ! if multiple of 4 limbs, skip first loop + mov %g1,%l1 + + sub %i2,%l4,%i2 ! adjust count for main loop + +.Loop0: ldx [%i1-16],%g3 + add %i0,-8,%i0 + add %i1,-8,%i1 + add %l4,-1,%l4 + sllx %g2,%i3,%i4 + srlx %g3,%i5,%g1 + mov %g3,%g2 + or %i4,%g1,%i4 + brnz,pt %l4,.Loop0 + stx %i4,[%i0+0] + +.L0: brz,pn %i2,.Lend + nop + +.Loop: ldx [%i1-16],%g3 + add %i0,-32,%i0 + add %i2,-4,%i2 + sllx %g2,%i3,%i4 + srlx %g3,%i5,%g1 + + ldx [%i1-24],%g2 + sllx %g3,%i3,%l4 + or %i4,%g1,%i4 + stx %i4,[%i0+24] + srlx %g2,%i5,%g1 + + ldx [%i1-32],%g3 + sllx %g2,%i3,%i4 + or %l4,%g1,%l4 + stx %l4,[%i0+16] + srlx %g3,%i5,%g1 + + ldx [%i1-40],%g2 + sllx %g3,%i3,%l4 + or %i4,%g1,%i4 + stx %i4,[%i0+8] + srlx %g2,%i5,%g1 + + add %i1,-32,%i1 + or %l4,%g1,%l4 + brnz,pt %i2,.Loop + stx %l4,[%i0+0] + +.Lend: sllx %g2,%i3,%g2 + stx %g2,[%i0-8] + + mov %l1,%i0 + jmpl %i7+8, %g0 + restore + +END(__mpn_lshift) diff --git a/sysdeps/sparc/sparc64/mul_1.S b/sysdeps/sparc/sparc64/mul_1.S new file mode 100644 index 0000000000..87e441d7eb --- /dev/null +++ b/sysdeps/sparc/sparc64/mul_1.S @@ -0,0 +1,84 @@ +/* SPARC v9 __mpn_mul_1 -- Multiply a limb vector with a single limb and + store the product in a second limb vector. + + Copyright (C) 1995, 1996 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + +#include + + +/* INPUT PARAMETERS + res_ptr o0 + s1_ptr o1 + size o2 + s2_limb o3 */ + +ENTRY(__mpn_mul_1) + !#PROLOGUE# 0 + save %sp,-128,%sp + !#PROLOGUE# 1 + + sub %g0,%i2,%o7 + sllx %o7,3,%g5 + sub %i1,%g5,%o3 + sub %i0,%g5,%o4 + mov 0,%o0 ! zero cy_limb + + srl %i3,0,%o1 ! extract low 32 bits of s2_limb + srlx %i3,32,%i3 ! extract high 32 bits of s2_limb + mov 1,%o2 + sllx %o2,32,%o2 ! o2 = 0x100000000 + + ! hi ! + ! mid-1 ! + ! mid-2 ! + ! lo ! +.Loop: + sllx %o7,3,%g1 + ldx [%o3+%g1],%g5 + srl %g5,0,%i0 ! zero hi bits + srlx %g5,32,%g5 + mulx %o1,%i0,%i4 ! lo product + mulx %i3,%i0,%i1 ! mid-1 product + mulx %o1,%g5,%l2 ! mid-2 product + mulx %i3,%g5,%i5 ! hi product + srlx %i4,32,%i0 ! extract high 32 bits of lo product... + add %i1,%i0,%i1 ! ...and add it to the mid-1 product + addcc %i1,%l2,%i1 ! add mid products + mov 0,%l0 ! we need the carry from that add... + movcs %xcc,%o2,%l0 ! ...compute it and... + add %i5,%l0,%i5 ! ...add to bit 32 of the hi product + sllx %i1,32,%i0 ! align low bits of mid product + srl %i4,0,%g5 ! zero high 32 bits of lo product + add %i0,%g5,%i0 ! combine into low 64 bits of result + srlx %i1,32,%i1 ! extract high bits of mid product... + add %i5,%i1,%i1 ! ...and add them to the high result + addcc %i0,%o0,%i0 ! add cy_limb to low 64 bits of result + mov 0,%g5 + movcs %xcc,1,%g5 + add %o7,1,%o7 + stx %i0,[%o4+%g1] + brnz %o7,.Loop + add %i1,%g5,%o0 ! compute new cy_limb + + mov %o0,%i0 + jmpl %i7+8,%g0 + restore + +END(__mpn_mul_1) diff --git a/sysdeps/sparc/sparc64/rshift.S b/sysdeps/sparc/sparc64/rshift.S new file mode 100644 index 0000000000..e1b3aca112 --- /dev/null +++ b/sysdeps/sparc/sparc64/rshift.S @@ -0,0 +1,93 @@ +/* SPARC v9 __mpn_rshift -- + + Copyright (C) 1996 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + +#include + +/* INPUT PARAMETERS + res_ptr %i0 + src_ptr %i1 + size %i2 + cnt %i3 */ + +ENTRY(__mpn_rshift) + save %sp, -128, %sp + + ldx [%i1],%g2 ! load first limb + sub %g0,%i3,%i5 ! negate shift count + add %i2,-1,%i2 + and %i2,4-1,%l4 ! number of limbs in first loop + sllx %g2,%i5,%g1 ! compute function result + brz,pn %l4,.L0 ! if multiple of 4 limbs, skip first loop + mov %g1,%l1 + + sub %i2,%l4,%i2 ! adjust count for main loop + +.Loop0: ldx [%i1+8],%g3 + add %i0,8,%i0 + add %i1,8,%i1 + add %l4,-1,%l4 + srlx %g2,%i3,%i4 + sllx %g3,%i5,%g1 + mov %g3,%g2 + or %i4,%g1,%i4 + brnz,pt %l4,.Loop0 + stx %i4,[%i0-8] + +.L0: brz,pn %i2,.Lend + nop + +.Loop: ldx [%i1+8],%g3 + add %i0,32,%i0 + add %i2,-4,%i2 + srlx %g2,%i3,%i4 + sllx %g3,%i5,%g1 + + ldx [%i1+16],%g2 + srlx %g3,%i3,%l4 + or %i4,%g1,%i4 + stx %i4,[%i0-32] + sllx %g2,%i5,%g1 + + ldx [%i1+24],%g3 + srlx %g2,%i3,%i4 + or %l4,%g1,%l4 + stx %l4,[%i0-24] + sllx %g3,%i5,%g1 + + ldx [%i1+32],%g2 + srlx %g3,%i3,%l4 + or %i4,%g1,%i4 + stx %i4,[%i0-16] + sllx %g2,%i5,%g1 + + add %i1,32,%i1 + or %l4,%g1,%l4 + brnz %i2,.Loop + stx %l4,[%i0-8] + +.Lend: srlx %g2,%i3,%g2 + stx %g2,[%i0-0] + + mov %l1,%i0 + jmpl %i7+8,%g0 + restore + +END(__mpn_rshift) diff --git a/sysdeps/sparc/sparc64/sub_n.S b/sysdeps/sparc/sparc64/sub_n.S new file mode 100644 index 0000000000..93f8a8235d --- /dev/null +++ b/sysdeps/sparc/sparc64/sub_n.S @@ -0,0 +1,55 @@ +/* SPARC v9 __mpn_sub_n -- Subtract two limb vectors of the same length > 0 + and store difference in a third limb vector. + + Copyright (C) 1995, 1996 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + +#include + +/* INPUT PARAMETERS + res_ptr %o0 + s1_ptr %o1 + s2_ptr %o2 + size %o3 */ + +ENTRY(__mpn_sub_n) + + sub %g0,%o3,%g3 + sllx %o3,3,%g1 + add %o1,%g1,%o1 ! make s1_ptr point at end + add %o2,%g1,%o2 ! make s2_ptr point at end + add %o0,%g1,%o0 ! make res_ptr point at end + mov 0,%o4 ! clear carry variable + sllx %g3,3,%o5 ! compute initial address index + +1: ldx [%o2+%o5],%g1 ! load s2 limb + add %g3,1,%g3 ! increment loop count + ldx [%o1+%o5],%g2 ! load s1 limb + addcc %g1,%o4,%g1 ! add s2 limb and carry variable + movcc %xcc,0,%o4 ! if carry-out, o4 was 1; clear it + subcc %g1,%g2,%g1 ! subtract s1 limb from sum + stx %g1,[%o0+%o5] ! store result + add %o5,8,%o5 ! increment address index + brnz,pt %g3,1b + movcs %xcc,1,%o4 ! if s1 subtract gave carry, record it + + retl + mov %o4,%o0 + +END(__mpn_sub_n) diff --git a/sysdeps/sparc/sparc64/submul_1.S b/sysdeps/sparc/sparc64/submul_1.S new file mode 100644 index 0000000000..3cc0e3bb08 --- /dev/null +++ b/sysdeps/sparc/sparc64/submul_1.S @@ -0,0 +1,87 @@ +/* SPARC v9 __mpn_submul_1 -- Multiply a limb vector with a single limb and + subtract the product from a second limb vector. + + Copyright (C) 1996 Free Software Foundation, Inc. + + This file is part of the GNU MP Library. + + The GNU MP Library is free software; you can redistribute it and/or modify + it under the terms of the GNU Library General Public License as published by + the Free Software Foundation; either version 2 of the License, or (at your + option) any later version. + + The GNU MP Library is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public + License for more details. + + You should have received a copy of the GNU Library General Public License + along with the GNU MP Library; see the file COPYING.LIB. If not, write to + the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, + MA 02111-1307, USA. */ + +#include + + +/* INPUT PARAMETERS + res_ptr o0 + s1_ptr o1 + size o2 + s2_limb o3 */ + +ENTRY(__mpn_submul_1) + !#PROLOGUE# 0 + save %sp,-128,%sp + !#PROLOGUE# 1 + + sub %g0,%i2,%o7 + sllx %o7,3,%g5 + sub %i1,%g5,%o3 + sub %i0,%g5,%o4 + mov 0,%o0 ! zero cy_limb + + srl %i3,0,%o1 ! extract low 32 bits of s2_limb + srlx %i3,32,%i3 ! extract high 32 bits of s2_limb + mov 1,%o2 + sllx %o2,32,%o2 ! o2 = 0x100000000 + + ! hi ! + ! mid-1 ! + ! mid-2 ! + ! lo ! +.Loop: + sllx %o7,3,%g1 + ldx [%o3+%g1],%g5 + srl %g5,0,%i0 ! zero hi bits + srlx %g5,32,%g5 + mulx %o1,%i0,%i4 ! lo product + mulx %i3,%i0,%i1 ! mid-1 product + mulx %o1,%g5,%l2 ! mid-2 product + mulx %i3,%g5,%i5 ! hi product + srlx %i4,32,%i0 ! extract high 32 bits of lo product... + add %i1,%i0,%i1 ! ...and add it to the mid-1 product + addcc %i1,%l2,%i1 ! add mid products + mov 0,%l0 ! we need the carry from that add... + movcs %xcc,%o2,%l0 ! ...compute it and... + add %i5,%l0,%i5 ! ...add to bit 32 of the hi product + sllx %i1,32,%i0 ! align low bits of mid product + srl %i4,0,%g5 ! zero high 32 bits of lo product + add %i0,%g5,%i0 ! combine into low 64 bits of result + srlx %i1,32,%i1 ! extract high bits of mid product... + add %i5,%i1,%i1 ! ...and add them to the high result + addcc %i0,%o0,%i0 ! add cy_limb to low 64 bits of result + mov 0,%g5 + movcs %xcc,1,%g5 + add %o7,1,%o7 + ldx [%o4+%g1],%l1 + subcc %l1,%i0,%i0 + movcs %xcc,1,%g5 + stx %i0,[%o4+%g1] + brnz %o7,.Loop + add %i1,%g5,%o0 ! compute new cy_limb + + mov %o0,%i0 + jmpl %i7+8,%g0 + restore + +END(__mpn_submul_1) diff --git a/sysdeps/sparc/sparc8/addmul_1.S b/sysdeps/sparc/sparc8/addmul_1.S deleted file mode 100644 index fb9ea7cf0e..0000000000 --- a/sysdeps/sparc/sparc8/addmul_1.S +++ /dev/null @@ -1,124 +0,0 @@ -! SPARC v8 __mpn_addmul_1 -- Multiply a limb vector with a limb and -! add the result to a second limb vector. - -! Copyright (C) 1992, 1993, 1994, 1995 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr o0 -! s1_ptr o1 -! size o2 -! s2_limb o3 - -#include "sysdep.h" - -.text - .align 4 - .global C_SYMBOL_NAME(__mpn_addmul_1) -C_SYMBOL_NAME(__mpn_addmul_1): - orcc %g0,%g0,%g2 - ld [%o1+0],%o4 ! 1 - - sll %o2,4,%g1 - and %g1,(4-1)<<4,%g1 -#if PIC - mov %o7,%g4 ! Save return address register - call 1f - add %o7,LL-1f,%g3 -1: mov %g4,%o7 ! Restore return address register -#else - sethi %hi(LL),%g3 - or %g3,%lo(LL),%g3 -#endif - jmp %g3+%g1 - nop -LL: -LL00: add %o0,-4,%o0 - b Loop00 /* 4, 8, 12, ... */ - add %o1,-4,%o1 - nop -LL01: b Loop01 /* 1, 5, 9, ... */ - nop - nop - nop -LL10: add %o0,-12,%o0 /* 2, 6, 10, ... */ - b Loop10 - add %o1,4,%o1 - nop -LL11: add %o0,-8,%o0 /* 3, 7, 11, ... */ - b Loop11 - add %o1,-8,%o1 - nop - -1: addcc %g3,%g2,%g3 ! 1 - ld [%o1+4],%o4 ! 2 - rd %y,%g2 ! 1 - addx %g0,%g2,%g2 - ld [%o0+0],%g1 ! 2 - addcc %g1,%g3,%g3 - st %g3,[%o0+0] ! 1 -Loop00: umul %o4,%o3,%g3 ! 2 - ld [%o0+4],%g1 ! 2 - addxcc %g3,%g2,%g3 ! 2 - ld [%o1+8],%o4 ! 3 - rd %y,%g2 ! 2 - addx %g0,%g2,%g2 - nop - addcc %g1,%g3,%g3 - st %g3,[%o0+4] ! 2 -Loop11: umul %o4,%o3,%g3 ! 3 - addxcc %g3,%g2,%g3 ! 3 - ld [%o1+12],%o4 ! 4 - rd %y,%g2 ! 3 - add %o1,16,%o1 - addx %g0,%g2,%g2 - ld [%o0+8],%g1 ! 2 - addcc %g1,%g3,%g3 - st %g3,[%o0+8] ! 3 -Loop10: umul %o4,%o3,%g3 ! 4 - addxcc %g3,%g2,%g3 ! 4 - ld [%o1+0],%o4 ! 1 - rd %y,%g2 ! 4 - addx %g0,%g2,%g2 - ld [%o0+12],%g1 ! 2 - addcc %g1,%g3,%g3 - st %g3,[%o0+12] ! 4 - add %o0,16,%o0 - addx %g0,%g2,%g2 -Loop01: addcc %o2,-4,%o2 - bg 1b - umul %o4,%o3,%g3 ! 1 - - addcc %g3,%g2,%g3 ! 4 - rd %y,%g2 ! 4 - addx %g0,%g2,%g2 - ld [%o0+0],%g1 ! 2 - addcc %g1,%g3,%g3 - st %g3,[%o0+0] ! 4 - addx %g0,%g2,%o0 - - retl - nop - - -! umul, ld, addxcc, rd, st - -! umul, ld, addxcc, rd, ld, addcc, st, addx - diff --git a/sysdeps/sparc/sparc8/mul_1.S b/sysdeps/sparc/sparc8/mul_1.S deleted file mode 100644 index b641feb453..0000000000 --- a/sysdeps/sparc/sparc8/mul_1.S +++ /dev/null @@ -1,99 +0,0 @@ -! SPARC v8 __mpn_mul_1 -- Multiply a limb vector with a single limb and -! store the product in a second limb vector. - -! Copyright (C) 1992, 1994, 1995 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr o0 -! s1_ptr o1 -! size o2 -! s2_limb o3 - -#include "sysdep.h" - -.text - .align 8 - .global C_SYMBOL_NAME(__mpn_mul_1) -C_SYMBOL_NAME(__mpn_mul_1): - sll %o2,4,%g1 - and %g1,(4-1)<<4,%g1 -#if PIC - mov %o7,%g4 ! Save return address register - call 1f - add %o7,LL-1f,%g3 -1: mov %g4,%o7 ! Restore return address register -#else - sethi %hi(LL),%g3 - or %g3,%lo(LL),%g3 -#endif - jmp %g3+%g1 - ld [%o1+0],%o4 ! 1 -LL: -LL00: add %o0,-4,%o0 - add %o1,-4,%o1 - b Loop00 /* 4, 8, 12, ... */ - orcc %g0,%g0,%g2 -LL01: b Loop01 /* 1, 5, 9, ... */ - orcc %g0,%g0,%g2 - nop - nop -LL10: add %o0,-12,%o0 /* 2, 6, 10, ... */ - add %o1,4,%o1 - b Loop10 - orcc %g0,%g0,%g2 - nop -LL11: add %o0,-8,%o0 /* 3, 7, 11, ... */ - add %o1,-8,%o1 - b Loop11 - orcc %g0,%g0,%g2 - -Loop: addcc %g3,%g2,%g3 ! 1 - ld [%o1+4],%o4 ! 2 - st %g3,[%o0+0] ! 1 - rd %y,%g2 ! 1 -Loop00: umul %o4,%o3,%g3 ! 2 - addxcc %g3,%g2,%g3 ! 2 - ld [%o1+8],%o4 ! 3 - st %g3,[%o0+4] ! 2 - rd %y,%g2 ! 2 -Loop11: umul %o4,%o3,%g3 ! 3 - addxcc %g3,%g2,%g3 ! 3 - ld [%o1+12],%o4 ! 4 - add %o1,16,%o1 - st %g3,[%o0+8] ! 3 - rd %y,%g2 ! 3 -Loop10: umul %o4,%o3,%g3 ! 4 - addxcc %g3,%g2,%g3 ! 4 - ld [%o1+0],%o4 ! 1 - st %g3,[%o0+12] ! 4 - add %o0,16,%o0 - rd %y,%g2 ! 4 - addx %g0,%g2,%g2 -Loop01: addcc %o2,-4,%o2 - bg Loop - umul %o4,%o3,%g3 ! 1 - - addcc %g3,%g2,%g3 ! 4 - st %g3,[%o0+0] ! 4 - rd %y,%g2 ! 4 - - retl - addx %g0,%g2,%o0 diff --git a/sysdeps/sparc/sparc8/submul_1.S b/sysdeps/sparc/sparc8/submul_1.S deleted file mode 100644 index e40119d011..0000000000 --- a/sysdeps/sparc/sparc8/submul_1.S +++ /dev/null @@ -1,58 +0,0 @@ -! SPARC v8 __mpn_submul_1 -- Multiply a limb vector with a limb and -! subtract the result from a second limb vector. - -! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr o0 -! s1_ptr o1 -! size o2 -! s2_limb o3 - -#include "sysdep.h" - -.text - .align 4 - .global C_SYMBOL_NAME(__mpn_submul_1) -C_SYMBOL_NAME(__mpn_submul_1): - sub %g0,%o2,%o2 ! negate ... - sll %o2,2,%o2 ! ... and scale size - sub %o1,%o2,%o1 ! o1 is offset s1_ptr - sub %o0,%o2,%g1 ! g1 is offset res_ptr - - mov 0,%o0 ! clear cy_limb - -Loop: ld [%o1+%o2],%o4 - ld [%g1+%o2],%g2 - umul %o4,%o3,%o5 - rd %y,%g3 - addcc %o5,%o0,%o5 - addx %g3,0,%o0 - subcc %g2,%o5,%g2 - addx %o0,0,%o0 - st %g2,[%g1+%o2] - - addcc %o2,4,%o2 - bne Loop - nop - - retl - nop diff --git a/sysdeps/sparc/sparc8/udiv_qrnnd.S b/sysdeps/sparc/sparc8/udiv_qrnnd.S deleted file mode 100644 index 49c2398806..0000000000 --- a/sysdeps/sparc/sparc8/udiv_qrnnd.S +++ /dev/null @@ -1,186 +0,0 @@ -! SPARC __udiv_qrnnd division support, used from longlong.h. - -! Copyright (C) 1993, 1994 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. - - -! INPUT PARAMETERS -! rem_ptr o0 -! n1 o1 -! n0 o2 -! d o3 - -#include "sysdep.h" - - .text - .align 4 - .global C_SYMBOL_NAME(__udiv_qrnnd) -C_SYMBOL_NAME(__udiv_qrnnd): - tst %o3 - bneg Largedivisor - mov 8,%g1 - - b Lp1 - addxcc %o2,%o2,%o2 - -Lplop: bcc Ln1 - addxcc %o2,%o2,%o2 -Lp1: addx %o1,%o1,%o1 - subcc %o1,%o3,%o4 - bcc Ln2 - addxcc %o2,%o2,%o2 -Lp2: addx %o1,%o1,%o1 - subcc %o1,%o3,%o4 - bcc Ln3 - addxcc %o2,%o2,%o2 -Lp3: addx %o1,%o1,%o1 - subcc %o1,%o3,%o4 - bcc Ln4 - addxcc %o2,%o2,%o2 -Lp4: addx %o1,%o1,%o1 - addcc %g1,-1,%g1 - bne Lplop - subcc %o1,%o3,%o4 - bcc Ln5 - addxcc %o2,%o2,%o2 -Lp5: st %o1,[%o0] - retl - xnor %g0,%o2,%o0 - -Lnlop: bcc Lp1 - addxcc %o2,%o2,%o2 -Ln1: addx %o4,%o4,%o4 - subcc %o4,%o3,%o1 - bcc Lp2 - addxcc %o2,%o2,%o2 -Ln2: addx %o4,%o4,%o4 - subcc %o4,%o3,%o1 - bcc Lp3 - addxcc %o2,%o2,%o2 -Ln3: addx %o4,%o4,%o4 - subcc %o4,%o3,%o1 - bcc Lp4 - addxcc %o2,%o2,%o2 -Ln4: addx %o4,%o4,%o4 - addcc %g1,-1,%g1 - bne Lnlop - subcc %o4,%o3,%o1 - bcc Lp5 - addxcc %o2,%o2,%o2 -Ln5: st %o4,[%o0] - retl - xnor %g0,%o2,%o0 - -Largedivisor: - and %o2,1,%o5 ! %o5 = n0 & 1 - - srl %o2,1,%o2 - sll %o1,31,%g2 - or %g2,%o2,%o2 ! %o2 = lo(n1n0 >> 1) - srl %o1,1,%o1 ! %o1 = hi(n1n0 >> 1) - - and %o3,1,%g2 - srl %o3,1,%g3 ! %g3 = floor(d / 2) - add %g3,%g2,%g3 ! %g3 = ceil(d / 2) - - b LLp1 - addxcc %o2,%o2,%o2 - -LLplop: bcc LLn1 - addxcc %o2,%o2,%o2 -LLp1: addx %o1,%o1,%o1 - subcc %o1,%g3,%o4 - bcc LLn2 - addxcc %o2,%o2,%o2 -LLp2: addx %o1,%o1,%o1 - subcc %o1,%g3,%o4 - bcc LLn3 - addxcc %o2,%o2,%o2 -LLp3: addx %o1,%o1,%o1 - subcc %o1,%g3,%o4 - bcc LLn4 - addxcc %o2,%o2,%o2 -LLp4: addx %o1,%o1,%o1 - addcc %g1,-1,%g1 - bne LLplop - subcc %o1,%g3,%o4 - bcc LLn5 - addxcc %o2,%o2,%o2 -LLp5: add %o1,%o1,%o1 ! << 1 - tst %g2 - bne Oddp - add %o5,%o1,%o1 - st %o1,[%o0] - retl - xnor %g0,%o2,%o0 - -LLnlop: bcc LLp1 - addxcc %o2,%o2,%o2 -LLn1: addx %o4,%o4,%o4 - subcc %o4,%g3,%o1 - bcc LLp2 - addxcc %o2,%o2,%o2 -LLn2: addx %o4,%o4,%o4 - subcc %o4,%g3,%o1 - bcc LLp3 - addxcc %o2,%o2,%o2 -LLn3: addx %o4,%o4,%o4 - subcc %o4,%g3,%o1 - bcc LLp4 - addxcc %o2,%o2,%o2 -LLn4: addx %o4,%o4,%o4 - addcc %g1,-1,%g1 - bne LLnlop - subcc %o4,%g3,%o1 - bcc LLp5 - addxcc %o2,%o2,%o2 -LLn5: add %o4,%o4,%o4 ! << 1 - tst %g2 - bne Oddn - add %o5,%o4,%o4 - st %o4,[%o0] - retl - xnor %g0,%o2,%o0 - -Oddp: xnor %g0,%o2,%o2 - ! q' in %o2. r' in %o1 - addcc %o1,%o2,%o1 - bcc LLp6 - addx %o2,0,%o2 - sub %o1,%o3,%o1 -LLp6: subcc %o1,%o3,%g0 - bcs LLp7 - subx %o2,-1,%o2 - sub %o1,%o3,%o1 -LLp7: st %o1,[%o0] - retl - mov %o2,%o0 - -Oddn: xnor %g0,%o2,%o2 - ! q' in %o2. r' in %o4 - addcc %o4,%o2,%o4 - bcc LLn6 - addx %o2,0,%o2 - sub %o4,%o3,%o4 -LLn6: subcc %o4,%o3,%g0 - bcs LLn7 - subx %o2,-1,%o2 - sub %o4,%o3,%o4 -LLn7: st %o4,[%o0] - retl - mov %o2,%o0 diff --git a/sysdeps/sparc/sub_n.S b/sysdeps/sparc/sub_n.S deleted file mode 100644 index b7a11958e2..0000000000 --- a/sysdeps/sparc/sub_n.S +++ /dev/null @@ -1,311 +0,0 @@ -! SPARC __mpn_sub_n -- Subtract two limb vectors of the same length > 0 and -! store difference in a third limb vector. - -! Copyright (C) 1995, 1996 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -#define res_ptr %o0 -#define s1_ptr %o1 -#define s2_ptr %o2 -#define size %o3 - -#include "sysdep.h" - - .text - .align 4 - .global C_SYMBOL_NAME(__mpn_sub_n) -C_SYMBOL_NAME(__mpn_sub_n): - xor s2_ptr,res_ptr,%g1 - andcc %g1,4,%g0 - bne L1 ! branch if alignment differs - nop -! ** V1a ** - andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 - be L_v1 ! if no, branch - nop -/* Add least significant limb separately to align res_ptr and s2_ptr */ - ld [s1_ptr],%g4 - add s1_ptr,4,s1_ptr - ld [s2_ptr],%g2 - add s2_ptr,4,s2_ptr - add size,-1,size - subcc %g4,%g2,%o4 - st %o4,[res_ptr] - add res_ptr,4,res_ptr -L_v1: addx %g0,%g0,%o4 ! save cy in register - cmp size,2 ! if size < 2 ... - bl Lend2 ! ... branch to tail code - subcc %g0,%o4,%g0 ! restore cy - - ld [s1_ptr+0],%g4 - addcc size,-10,size - ld [s1_ptr+4],%g1 - ldd [s2_ptr+0],%g2 - blt Lfin1 - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 8 limbs until less than 8 limbs remain */ -Loop1: subxcc %g4,%g2,%o4 - ld [s1_ptr+8],%g4 - subxcc %g1,%g3,%o5 - ld [s1_ptr+12],%g1 - ldd [s2_ptr+8],%g2 - std %o4,[res_ptr+0] - subxcc %g4,%g2,%o4 - ld [s1_ptr+16],%g4 - subxcc %g1,%g3,%o5 - ld [s1_ptr+20],%g1 - ldd [s2_ptr+16],%g2 - std %o4,[res_ptr+8] - subxcc %g4,%g2,%o4 - ld [s1_ptr+24],%g4 - subxcc %g1,%g3,%o5 - ld [s1_ptr+28],%g1 - ldd [s2_ptr+24],%g2 - std %o4,[res_ptr+16] - subxcc %g4,%g2,%o4 - ld [s1_ptr+32],%g4 - subxcc %g1,%g3,%o5 - ld [s1_ptr+36],%g1 - ldd [s2_ptr+32],%g2 - std %o4,[res_ptr+24] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - add s1_ptr,32,s1_ptr - add s2_ptr,32,s2_ptr - add res_ptr,32,res_ptr - bge Loop1 - subcc %g0,%o4,%g0 ! restore cy - -Lfin1: addcc size,8-2,size - blt Lend1 - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 2 limbs until less than 2 limbs remain */ -Loope1: subxcc %g4,%g2,%o4 - ld [s1_ptr+8],%g4 - subxcc %g1,%g3,%o5 - ld [s1_ptr+12],%g1 - ldd [s2_ptr+8],%g2 - std %o4,[res_ptr+0] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-2,size - add s1_ptr,8,s1_ptr - add s2_ptr,8,s2_ptr - add res_ptr,8,res_ptr - bge Loope1 - subcc %g0,%o4,%g0 ! restore cy -Lend1: subxcc %g4,%g2,%o4 - subxcc %g1,%g3,%o5 - std %o4,[res_ptr+0] - addx %g0,%g0,%o4 ! save cy in register - - andcc size,1,%g0 - be Lret1 - subcc %g0,%o4,%g0 ! restore cy -/* Add last limb */ - ld [s1_ptr+8],%g4 - ld [s2_ptr+8],%g2 - subxcc %g4,%g2,%o4 - st %o4,[res_ptr+8] - -Lret1: retl - addx %g0,%g0,%o0 ! return carry-out from most sign. limb - -L1: xor s1_ptr,res_ptr,%g1 - andcc %g1,4,%g0 - bne L2 - nop -! ** V1b ** - andcc res_ptr,4,%g0 ! res_ptr unaligned? Side effect: cy=0 - be L_v1b ! if no, branch - nop -/* Add least significant limb separately to align res_ptr and s1_ptr */ - ld [s2_ptr],%g4 - add s2_ptr,4,s2_ptr - ld [s1_ptr],%g2 - add s1_ptr,4,s1_ptr - add size,-1,size - subcc %g2,%g4,%o4 - st %o4,[res_ptr] - add res_ptr,4,res_ptr -L_v1b: addx %g0,%g0,%o4 ! save cy in register - cmp size,2 ! if size < 2 ... - bl Lend2 ! ... branch to tail code - subcc %g0,%o4,%g0 ! restore cy - - ld [s2_ptr+0],%g4 - addcc size,-10,size - ld [s2_ptr+4],%g1 - ldd [s1_ptr+0],%g2 - blt Lfin1b - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 8 limbs until less than 8 limbs remain */ -Loop1b: subxcc %g2,%g4,%o4 - ld [s2_ptr+8],%g4 - subxcc %g3,%g1,%o5 - ld [s2_ptr+12],%g1 - ldd [s1_ptr+8],%g2 - std %o4,[res_ptr+0] - subxcc %g2,%g4,%o4 - ld [s2_ptr+16],%g4 - subxcc %g3,%g1,%o5 - ld [s2_ptr+20],%g1 - ldd [s1_ptr+16],%g2 - std %o4,[res_ptr+8] - subxcc %g2,%g4,%o4 - ld [s2_ptr+24],%g4 - subxcc %g3,%g1,%o5 - ld [s2_ptr+28],%g1 - ldd [s1_ptr+24],%g2 - std %o4,[res_ptr+16] - subxcc %g2,%g4,%o4 - ld [s2_ptr+32],%g4 - subxcc %g3,%g1,%o5 - ld [s2_ptr+36],%g1 - ldd [s1_ptr+32],%g2 - std %o4,[res_ptr+24] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - add s1_ptr,32,s1_ptr - add s2_ptr,32,s2_ptr - add res_ptr,32,res_ptr - bge Loop1b - subcc %g0,%o4,%g0 ! restore cy - -Lfin1b: addcc size,8-2,size - blt Lend1b - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 2 limbs until less than 2 limbs remain */ -Loope1b:subxcc %g2,%g4,%o4 - ld [s2_ptr+8],%g4 - subxcc %g3,%g1,%o5 - ld [s2_ptr+12],%g1 - ldd [s1_ptr+8],%g2 - std %o4,[res_ptr+0] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-2,size - add s1_ptr,8,s1_ptr - add s2_ptr,8,s2_ptr - add res_ptr,8,res_ptr - bge Loope1b - subcc %g0,%o4,%g0 ! restore cy -Lend1b: subxcc %g2,%g4,%o4 - subxcc %g3,%g1,%o5 - std %o4,[res_ptr+0] - addx %g0,%g0,%o4 ! save cy in register - - andcc size,1,%g0 - be Lret1b - subcc %g0,%o4,%g0 ! restore cy -/* Add last limb */ - ld [s2_ptr+8],%g4 - ld [s1_ptr+8],%g2 - subxcc %g2,%g4,%o4 - st %o4,[res_ptr+8] - -Lret1b: retl - addx %g0,%g0,%o0 ! return carry-out from most sign. limb - -! ** V2 ** -/* If we come here, the alignment of s1_ptr and res_ptr as well as the - alignment of s2_ptr and res_ptr differ. Since there are only two ways - things can be aligned (that we care about) we now know that the alignment - of s1_ptr and s2_ptr are the same. */ - -L2: cmp size,1 - be Ljone - nop - andcc s1_ptr,4,%g0 ! s1_ptr unaligned? Side effect: cy=0 - be L_v2 ! if no, branch - nop -/* Add least significant limb separately to align s1_ptr and s2_ptr */ - ld [s1_ptr],%g4 - add s1_ptr,4,s1_ptr - ld [s2_ptr],%g2 - add s2_ptr,4,s2_ptr - add size,-1,size - subcc %g4,%g2,%o4 - st %o4,[res_ptr] - add res_ptr,4,res_ptr - -L_v2: addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - blt Lfin2 - subcc %g0,%o4,%g0 ! restore cy -/* Add blocks of 8 limbs until less than 8 limbs remain */ -Loop2: ldd [s1_ptr+0],%g2 - ldd [s2_ptr+0],%o4 - subxcc %g2,%o4,%g2 - st %g2,[res_ptr+0] - subxcc %g3,%o5,%g3 - st %g3,[res_ptr+4] - ldd [s1_ptr+8],%g2 - ldd [s2_ptr+8],%o4 - subxcc %g2,%o4,%g2 - st %g2,[res_ptr+8] - subxcc %g3,%o5,%g3 - st %g3,[res_ptr+12] - ldd [s1_ptr+16],%g2 - ldd [s2_ptr+16],%o4 - subxcc %g2,%o4,%g2 - st %g2,[res_ptr+16] - subxcc %g3,%o5,%g3 - st %g3,[res_ptr+20] - ldd [s1_ptr+24],%g2 - ldd [s2_ptr+24],%o4 - subxcc %g2,%o4,%g2 - st %g2,[res_ptr+24] - subxcc %g3,%o5,%g3 - st %g3,[res_ptr+28] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-8,size - add s1_ptr,32,s1_ptr - add s2_ptr,32,s2_ptr - add res_ptr,32,res_ptr - bge Loop2 - subcc %g0,%o4,%g0 ! restore cy - -Lfin2: addcc size,8-2,size - blt Lend2 - subcc %g0,%o4,%g0 ! restore cy -Loope2: ldd [s1_ptr+0],%g2 - ldd [s2_ptr+0],%o4 - subxcc %g2,%o4,%g2 - st %g2,[res_ptr+0] - subxcc %g3,%o5,%g3 - st %g3,[res_ptr+4] - addx %g0,%g0,%o4 ! save cy in register - addcc size,-2,size - add s1_ptr,8,s1_ptr - add s2_ptr,8,s2_ptr - add res_ptr,8,res_ptr - bge Loope2 - subcc %g0,%o4,%g0 ! restore cy -Lend2: andcc size,1,%g0 - be Lret2 - subcc %g0,%o4,%g0 ! restore cy -/* Add last limb */ -Ljone: ld [s1_ptr],%g4 - ld [s2_ptr],%g2 - subxcc %g4,%g2,%o4 - st %o4,[res_ptr] - -Lret2: retl - addx %g0,%g0,%o0 ! return carry-out from most sign. limb diff --git a/sysdeps/sparc/submul_1.S b/sysdeps/sparc/submul_1.S deleted file mode 100644 index a8ebd501a7..0000000000 --- a/sysdeps/sparc/submul_1.S +++ /dev/null @@ -1,147 +0,0 @@ -! SPARC __mpn_submul_1 -- Multiply a limb vector with a limb and subtract -! the result from a second limb vector. - -! Copyright (C) 1992, 1993, 1994 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, -! MA 02111-1307, USA. - - -! INPUT PARAMETERS -! res_ptr o0 -! s1_ptr o1 -! size o2 -! s2_limb o3 - -#include "sysdep.h" - -.text - .align 4 - .global C_SYMBOL_NAME(__mpn_submul_1) -C_SYMBOL_NAME(__mpn_submul_1): - ! Make S1_PTR and RES_PTR point at the end of their blocks - ! and put (- 4 x SIZE) in index/loop counter. - sll %o2,2,%o2 - add %o0,%o2,%o4 ! RES_PTR in o4 since o0 is retval - add %o1,%o2,%o1 - sub %g0,%o2,%o2 - - cmp %o3,0xfff - bgu Large - nop - - ld [%o1+%o2],%o5 - mov 0,%o0 - b L0 - add %o4,-4,%o4 -Loop0: - subcc %o5,%g1,%g1 - ld [%o1+%o2],%o5 - addx %o0,%g0,%o0 - st %g1,[%o4+%o2] -L0: wr %g0,%o3,%y - sra %o5,31,%g2 - and %o3,%g2,%g2 - andcc %g1,0,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,%o5,%g1 - mulscc %g1,0,%g1 - sra %g1,20,%g4 - sll %g1,12,%g1 - rd %y,%g3 - srl %g3,20,%g3 - or %g1,%g3,%g1 - - addcc %g1,%o0,%g1 - addx %g2,%g4,%o0 ! add sign-compensation and cy to hi limb - addcc %o2,4,%o2 ! loop counter - bne Loop0 - ld [%o4+%o2],%o5 - - subcc %o5,%g1,%g1 - addx %o0,%g0,%o0 - retl - st %g1,[%o4+%o2] - - -Large: ld [%o1+%o2],%o5 - mov 0,%o0 - sra %o3,31,%g4 ! g4 = mask of ones iff S2_LIMB < 0 - b L1 - add %o4,-4,%o4 -Loop: - subcc %o5,%g3,%g3 - ld [%o1+%o2],%o5 - addx %o0,%g0,%o0 - st %g3,[%o4+%o2] -L1: wr %g0,%o5,%y - and %o5,%g4,%g2 - andcc %g0,%g0,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%o3,%g1 - mulscc %g1,%g0,%g1 - rd %y,%g3 - addcc %g3,%o0,%g3 - addx %g2,%g1,%o0 - addcc %o2,4,%o2 - bne Loop - ld [%o4+%o2],%o5 - - subcc %o5,%g3,%g3 - addx %o0,%g0,%o0 - retl - st %g3,[%o4+%o2] diff --git a/sysdeps/sparc/udiv.S b/sysdeps/sparc/udiv.S deleted file mode 100644 index 4a7d1526b0..0000000000 --- a/sysdeps/sparc/udiv.S +++ /dev/null @@ -1,352 +0,0 @@ - /* This file is generated from divrem.m4; DO NOT EDIT! */ -/* - * Division and remainder, from Appendix E of the Sparc Version 8 - * Architecture Manual, with fixes from Gordon Irlam. - */ - -/* - * Input: dividend and divisor in %o0 and %o1 respectively. - * - * m4 parameters: - * .udiv name of function to generate - * div div=div => %o0 / %o1; div=rem => %o0 % %o1 - * false false=true => signed; false=false => unsigned - * - * Algorithm parameters: - * N how many bits per iteration we try to get (4) - * WORDSIZE total number of bits (32) - * - * Derived constants: - * TOPBITS number of bits in the top decade of a number - * - * Important variables: - * Q the partial quotient under development (initially 0) - * R the remainder so far, initially the dividend - * ITER number of main division loop iterations required; - * equal to ceil(log2(quotient) / N). Note that this - * is the log base (2^N) of the quotient. - * V the current comparand, initially divisor*2^(ITER*N-1) - * - * Cost: - * Current estimate for non-large dividend is - * ceil(log2(quotient) / N) * (10 + 7N/2) + C - * A large dividend is one greater than 2^(31-TOPBITS) and takes a - * different path, as the upper bits of the quotient must be developed - * one bit at a time. - */ - - - -#include "sysdep.h" -#ifdef __linux__ -#include -#else -#ifdef __svr4__ -#include -#else -#include -#endif -#endif - -ENTRY(.udiv) - - ! Ready to divide. Compute size of quotient; scale comparand. - orcc %o1, %g0, %o5 - bne 1f - mov %o0, %o3 - - ! Divide by zero trap. If it returns, return 0 (about as - ! wrong as possible, but that is what SunOS does...). - ta ST_DIV0 - retl - clr %o0 - -1: - cmp %o3, %o5 ! if %o1 exceeds %o0, done - blu Lgot_result ! (and algorithm fails otherwise) - clr %o2 - sethi %hi(1 << (32 - 4 - 1)), %g1 - cmp %o3, %g1 - blu Lnot_really_big - clr %o4 - - ! Here the dividend is >= 2**(31-N) or so. We must be careful here, - ! as our usual N-at-a-shot divide step will cause overflow and havoc. - ! The number of bits in the result here is N*ITER+SC, where SC <= N. - ! Compute ITER in an unorthodox manner: know we need to shift V into - ! the top decade: so do not even bother to compare to R. - 1: - cmp %o5, %g1 - bgeu 3f - mov 1, %g7 - sll %o5, 4, %o5 - b 1b - add %o4, 1, %o4 - - ! Now compute %g7. - 2: addcc %o5, %o5, %o5 - bcc Lnot_too_big - add %g7, 1, %g7 - - ! We get here if the %o1 overflowed while shifting. - ! This means that %o3 has the high-order bit set. - ! Restore %o5 and subtract from %o3. - sll %g1, 4, %g1 ! high order bit - srl %o5, 1, %o5 ! rest of %o5 - add %o5, %g1, %o5 - b Ldo_single_div - sub %g7, 1, %g7 - - Lnot_too_big: - 3: cmp %o5, %o3 - blu 2b - nop - be Ldo_single_div - nop - /* NB: these are commented out in the V8-Sparc manual as well */ - /* (I do not understand this) */ - ! %o5 > %o3: went too far: back up 1 step - ! srl %o5, 1, %o5 - ! dec %g7 - ! do single-bit divide steps - ! - ! We have to be careful here. We know that %o3 >= %o5, so we can do the - ! first divide step without thinking. BUT, the others are conditional, - ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- - ! order bit set in the first step, just falling into the regular - ! division loop will mess up the first time around. - ! So we unroll slightly... - Ldo_single_div: - subcc %g7, 1, %g7 - bl Lend_regular_divide - nop - sub %o3, %o5, %o3 - mov 1, %o2 - b Lend_single_divloop - nop - Lsingle_divloop: - sll %o2, 1, %o2 - bl 1f - srl %o5, 1, %o5 - ! %o3 >= 0 - sub %o3, %o5, %o3 - b 2f - add %o2, 1, %o2 - 1: ! %o3 < 0 - add %o3, %o5, %o3 - sub %o2, 1, %o2 - 2: - Lend_single_divloop: - subcc %g7, 1, %g7 - bge Lsingle_divloop - tst %o3 - b,a Lend_regular_divide - -Lnot_really_big: -1: - sll %o5, 4, %o5 - cmp %o5, %o3 - bleu 1b - addcc %o4, 1, %o4 - be Lgot_result - sub %o4, 1, %o4 - - tst %o3 ! set up for initial iteration -Ldivloop: - sll %o2, 4, %o2 - ! depth 1, accumulated bits 0 - bl L.1.16 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 2, accumulated bits 1 - bl L.2.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits 3 - bl L.3.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 7 - bl L.4.23 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (7*2+1), %o2 - -L.4.23: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (7*2-1), %o2 - - -L.3.19: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 5 - bl L.4.21 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (5*2+1), %o2 - -L.4.21: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (5*2-1), %o2 - - - -L.2.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits 1 - bl L.3.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 3 - bl L.4.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (3*2+1), %o2 - -L.4.19: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (3*2-1), %o2 - - -L.3.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 1 - bl L.4.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (1*2+1), %o2 - -L.4.17: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (1*2-1), %o2 - - - - -L.1.16: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 2, accumulated bits -1 - bl L.2.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits -1 - bl L.3.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -1 - bl L.4.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2+1), %o2 - -L.4.15: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2-1), %o2 - - -L.3.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -3 - bl L.4.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2+1), %o2 - -L.4.13: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2-1), %o2 - - - -L.2.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits -3 - bl L.3.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -5 - bl L.4.11 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2+1), %o2 - -L.4.11: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2-1), %o2 - - -L.3.13: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -7 - bl L.4.9 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2+1), %o2 - -L.4.9: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2-1), %o2 - - - - - 9: -Lend_regular_divide: - subcc %o4, 1, %o4 - bge Ldivloop - tst %o3 - bl,a Lgot_result - ! non-restoring fixup here (one instruction only!) - sub %o2, 1, %o2 - - -Lgot_result: - - retl - mov %o2, %o0 diff --git a/sysdeps/sparc/udiv_qrnnd.S b/sysdeps/sparc/udiv_qrnnd.S deleted file mode 100644 index 437d66e9ea..0000000000 --- a/sysdeps/sparc/udiv_qrnnd.S +++ /dev/null @@ -1,164 +0,0 @@ -! SPARC __udiv_qrnnd division support, used from longlong.h. - -! Copyright (C) 1993, 1994, 1997 Free Software Foundation, Inc. - -! This file is part of the GNU MP Library. - -! The GNU MP Library is free software; you can redistribute it and/or modify -! it under the terms of the GNU Library General Public License as published by -! the Free Software Foundation; either version 2 of the License, or (at your -! option) any later version. - -! The GNU MP Library is distributed in the hope that it will be useful, but -! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -! or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public -! License for more details. - -! You should have received a copy of the GNU Library General Public License -! along with the GNU MP Library; see the file COPYING.LIB. If not, write to -! the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. - -! -! Added PIC support - May/96, Miguel de Icaza - -! INPUT PARAMETERS -! rem_ptr i0 -! n1 i1 -! n0 i2 -! d i3 - -#include "sysdep.h" -#undef ret /* Kludge for glibc */ - - .text - .align 8 -LC0: .double 0r4294967296 -LC1: .double 0r2147483648 - - .align 4 - .global __udiv_qrnnd - .type __udiv_qrnnd,@function -ENTRY(__udiv_qrnnd) - !#PROLOGUE# 0 - save %sp,-104,%sp - !#PROLOGUE# 1 - st %i1,[%fp-8] - ld [%fp-8],%f10 -#ifdef PIC -._XL11: - call ._XL1 - fitod %f10,%f4 -._XL1: - sub %o7,(._XL11-LC0),%o7 - ldd [%o7],%f8 -#else - sethi %hi(LC0),%o7 - fitod %f10,%f4 - ldd [%o7+%lo(LC0)],%f8 -#endif - cmp %i1,0 - bge L248 - mov %i0,%i5 - faddd %f4,%f8,%f4 -L248: - st %i2,[%fp-8] - ld [%fp-8],%f10 - fmuld %f4,%f8,%f6 - cmp %i2,0 - bge L249 - fitod %f10,%f2 - faddd %f2,%f8,%f2 -L249: - st %i3,[%fp-8] - faddd %f6,%f2,%f2 - ld [%fp-8],%f10 - cmp %i3,0 - bge L250 - fitod %f10,%f4 - faddd %f4,%f8,%f4 -L250: - fdivd %f2,%f4,%f2 -#ifdef PIC -._XL22: - call ._XL2 - nop -._XL2: - sub %o7,(._XL22-LC1),%o7 - ldd [%o7],%f4 -#else - sethi %hi(LC1),%o7 - ldd [%o7+%lo(LC1)],%f4 -#endif - fcmped %f2,%f4 - nop - fbge,a L251 - fsubd %f2,%f4,%f2 - fdtoi %f2,%f2 - st %f2,[%fp-8] - b L252 - ld [%fp-8],%i4 -L251: - fdtoi %f2,%f2 - st %f2,[%fp-8] - ld [%fp-8],%i4 - sethi %hi(-2147483648),%g2 - xor %i4,%g2,%i4 -L252: - wr %g0,%i4,%y - sra %i3,31,%g2 - and %i4,%g2,%g2 - andcc %g0,0,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,%i3,%g1 - mulscc %g1,0,%g1 - add %g1,%g2,%i0 - rd %y,%g3 - subcc %i2,%g3,%o7 - subxcc %i1,%i0,%g0 - be L253 - cmp %o7,%i3 - - add %i4,-1,%i0 - add %o7,%i3,%o7 - st %o7,[%i5] - ret - restore -L253: - blu L246 - mov %i4,%i0 - add %i4,1,%i0 - sub %o7,%i3,%o7 -L246: - st %o7,[%i5] - ret - restore diff --git a/sysdeps/sparc/umul.S b/sysdeps/sparc/umul.S deleted file mode 100644 index 7a26c295cb..0000000000 --- a/sysdeps/sparc/umul.S +++ /dev/null @@ -1,153 +0,0 @@ -/* - * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the - * upper 32 bits of the 64-bit product). - * - * This code optimizes short (less than 13-bit) multiplies. Short - * multiplies require 25 instruction cycles, and long ones require - * 45 instruction cycles. - * - * On return, overflow has occurred (%o1 is not zero) if and only if - * the Z condition code is clear, allowing, e.g., the following: - * - * call .umul - * nop - * bnz overflow (or tnz) - */ - -#include "DEFS.h" -FUNC(.umul) - or %o0, %o1, %o4 - mov %o0, %y ! multiplier -> Y - andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args - be Lmul_shortway ! if zero, can do it the short way - andcc %g0, %g0, %o4 ! zero the partial product and clear N and V - - /* - * Long multiply. 32 steps, followed by a final shift step. - */ - mulscc %o4, %o1, %o4 ! 1 - mulscc %o4, %o1, %o4 ! 2 - mulscc %o4, %o1, %o4 ! 3 - mulscc %o4, %o1, %o4 ! 4 - mulscc %o4, %o1, %o4 ! 5 - mulscc %o4, %o1, %o4 ! 6 - mulscc %o4, %o1, %o4 ! 7 - mulscc %o4, %o1, %o4 ! 8 - mulscc %o4, %o1, %o4 ! 9 - mulscc %o4, %o1, %o4 ! 10 - mulscc %o4, %o1, %o4 ! 11 - mulscc %o4, %o1, %o4 ! 12 - mulscc %o4, %o1, %o4 ! 13 - mulscc %o4, %o1, %o4 ! 14 - mulscc %o4, %o1, %o4 ! 15 - mulscc %o4, %o1, %o4 ! 16 - mulscc %o4, %o1, %o4 ! 17 - mulscc %o4, %o1, %o4 ! 18 - mulscc %o4, %o1, %o4 ! 19 - mulscc %o4, %o1, %o4 ! 20 - mulscc %o4, %o1, %o4 ! 21 - mulscc %o4, %o1, %o4 ! 22 - mulscc %o4, %o1, %o4 ! 23 - mulscc %o4, %o1, %o4 ! 24 - mulscc %o4, %o1, %o4 ! 25 - mulscc %o4, %o1, %o4 ! 26 - mulscc %o4, %o1, %o4 ! 27 - mulscc %o4, %o1, %o4 ! 28 - mulscc %o4, %o1, %o4 ! 29 - mulscc %o4, %o1, %o4 ! 30 - mulscc %o4, %o1, %o4 ! 31 - mulscc %o4, %o1, %o4 ! 32 - mulscc %o4, %g0, %o4 ! final shift - - - /* - * Normally, with the shift-and-add approach, if both numbers are - * positive you get the correct result. With 32-bit two's-complement - * numbers, -x is represented as - * - * x 32 - * ( 2 - ------ ) mod 2 * 2 - * 32 - * 2 - * - * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s, - * we can treat this as if the radix point were just to the left - * of the sign bit (multiply by 2^32), and get - * - * -x = (2 - x) mod 2 - * - * Then, ignoring the `mod 2's for convenience: - * - * x * y = xy - * -x * y = 2y - xy - * x * -y = 2x - xy - * -x * -y = 4 - 2x - 2y + xy - * - * For signed multiplies, we subtract (x << 32) from the partial - * product to fix this problem for negative multipliers (see mul.s). - * Because of the way the shift into the partial product is calculated - * (N xor V), this term is automatically removed for the multiplicand, - * so we don't have to adjust. - * - * But for unsigned multiplies, the high order bit wasn't a sign bit, - * and the correction is wrong. So for unsigned multiplies where the - * high order bit is one, we end up with xy - (y << 32). To fix it - * we add y << 32. - */ -#if 0 - tst %o1 - bl,a 1f ! if %o1 < 0 (high order bit = 1), - add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half) -1: rd %y, %o0 ! get lower half of product - retl - addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0 -#else - /* Faster code from tege@sics.se. */ - sra %o1, 31, %o2 ! make mask from sign bit - and %o0, %o2, %o2 ! %o2 = 0 or %o0, depending on sign of %o1 - rd %y, %o0 ! get lower half of product - retl - addcc %o4, %o2, %o1 ! add compensation and put upper half in place -#endif - -Lmul_shortway: - /* - * Short multiply. 12 steps, followed by a final shift step. - * The resulting bits are off by 12 and (32-12) = 20 bit positions, - * but there is no problem with %o0 being negative (unlike above), - * and overflow is impossible (the answer is at most 24 bits long). - */ - mulscc %o4, %o1, %o4 ! 1 - mulscc %o4, %o1, %o4 ! 2 - mulscc %o4, %o1, %o4 ! 3 - mulscc %o4, %o1, %o4 ! 4 - mulscc %o4, %o1, %o4 ! 5 - mulscc %o4, %o1, %o4 ! 6 - mulscc %o4, %o1, %o4 ! 7 - mulscc %o4, %o1, %o4 ! 8 - mulscc %o4, %o1, %o4 ! 9 - mulscc %o4, %o1, %o4 ! 10 - mulscc %o4, %o1, %o4 ! 11 - mulscc %o4, %o1, %o4 ! 12 - mulscc %o4, %g0, %o4 ! final shift - - /* - * %o4 has 20 of the bits that should be in the result; %y has - * the bottom 12 (as %y's top 12). That is: - * - * %o4 %y - * +----------------+----------------+ - * | -12- | -20- | -12- | -20- | - * +------(---------+------)---------+ - * -----result----- - * - * The 12 bits of %o4 left of the `result' area are all zero; - * in fact, all top 20 bits of %o4 are zero. - */ - - rd %y, %o5 - sll %o4, 12, %o0 ! shift middle bits left 12 - srl %o5, 20, %o5 ! shift low bits right 20 - or %o5, %o0, %o0 - retl - addcc %g0, %g0, %o1 ! %o1 = zero, and set Z diff --git a/sysdeps/sparc/urem.S b/sysdeps/sparc/urem.S deleted file mode 100644 index e72c33dd63..0000000000 --- a/sysdeps/sparc/urem.S +++ /dev/null @@ -1,352 +0,0 @@ - /* This file is generated from divrem.m4; DO NOT EDIT! */ -/* - * Division and remainder, from Appendix E of the Sparc Version 8 - * Architecture Manual, with fixes from Gordon Irlam. - */ - -/* - * Input: dividend and divisor in %o0 and %o1 respectively. - * - * m4 parameters: - * .urem name of function to generate - * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 - * false false=true => signed; false=false => unsigned - * - * Algorithm parameters: - * N how many bits per iteration we try to get (4) - * WORDSIZE total number of bits (32) - * - * Derived constants: - * TOPBITS number of bits in the top decade of a number - * - * Important variables: - * Q the partial quotient under development (initially 0) - * R the remainder so far, initially the dividend - * ITER number of main division loop iterations required; - * equal to ceil(log2(quotient) / N). Note that this - * is the log base (2^N) of the quotient. - * V the current comparand, initially divisor*2^(ITER*N-1) - * - * Cost: - * Current estimate for non-large dividend is - * ceil(log2(quotient) / N) * (10 + 7N/2) + C - * A large dividend is one greater than 2^(31-TOPBITS) and takes a - * different path, as the upper bits of the quotient must be developed - * one bit at a time. - */ - - - -#include "sysdep.h" -#ifdef __linux__ -#include -#else -#ifdef __svr4__ -#include -#else -#include -#endif -#endif - -ENTRY(.urem) - - ! Ready to divide. Compute size of quotient; scale comparand. - orcc %o1, %g0, %o5 - bne 1f - mov %o0, %o3 - - ! Divide by zero trap. If it returns, return 0 (about as - ! wrong as possible, but that is what SunOS does...). - ta ST_DIV0 - retl - clr %o0 - -1: - cmp %o3, %o5 ! if %o1 exceeds %o0, done - blu Lgot_result ! (and algorithm fails otherwise) - clr %o2 - sethi %hi(1 << (32 - 4 - 1)), %g1 - cmp %o3, %g1 - blu Lnot_really_big - clr %o4 - - ! Here the dividend is >= 2**(31-N) or so. We must be careful here, - ! as our usual N-at-a-shot divide step will cause overflow and havoc. - ! The number of bits in the result here is N*ITER+SC, where SC <= N. - ! Compute ITER in an unorthodox manner: know we need to shift V into - ! the top decade: so do not even bother to compare to R. - 1: - cmp %o5, %g1 - bgeu 3f - mov 1, %g7 - sll %o5, 4, %o5 - b 1b - add %o4, 1, %o4 - - ! Now compute %g7. - 2: addcc %o5, %o5, %o5 - bcc Lnot_too_big - add %g7, 1, %g7 - - ! We get here if the %o1 overflowed while shifting. - ! This means that %o3 has the high-order bit set. - ! Restore %o5 and subtract from %o3. - sll %g1, 4, %g1 ! high order bit - srl %o5, 1, %o5 ! rest of %o5 - add %o5, %g1, %o5 - b Ldo_single_div - sub %g7, 1, %g7 - - Lnot_too_big: - 3: cmp %o5, %o3 - blu 2b - nop - be Ldo_single_div - nop - /* NB: these are commented out in the V8-Sparc manual as well */ - /* (I do not understand this) */ - ! %o5 > %o3: went too far: back up 1 step - ! srl %o5, 1, %o5 - ! dec %g7 - ! do single-bit divide steps - ! - ! We have to be careful here. We know that %o3 >= %o5, so we can do the - ! first divide step without thinking. BUT, the others are conditional, - ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- - ! order bit set in the first step, just falling into the regular - ! division loop will mess up the first time around. - ! So we unroll slightly... - Ldo_single_div: - subcc %g7, 1, %g7 - bl Lend_regular_divide - nop - sub %o3, %o5, %o3 - mov 1, %o2 - b Lend_single_divloop - nop - Lsingle_divloop: - sll %o2, 1, %o2 - bl 1f - srl %o5, 1, %o5 - ! %o3 >= 0 - sub %o3, %o5, %o3 - b 2f - add %o2, 1, %o2 - 1: ! %o3 < 0 - add %o3, %o5, %o3 - sub %o2, 1, %o2 - 2: - Lend_single_divloop: - subcc %g7, 1, %g7 - bge Lsingle_divloop - tst %o3 - b,a Lend_regular_divide - -Lnot_really_big: -1: - sll %o5, 4, %o5 - cmp %o5, %o3 - bleu 1b - addcc %o4, 1, %o4 - be Lgot_result - sub %o4, 1, %o4 - - tst %o3 ! set up for initial iteration -Ldivloop: - sll %o2, 4, %o2 - ! depth 1, accumulated bits 0 - bl L.1.16 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 2, accumulated bits 1 - bl L.2.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits 3 - bl L.3.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 7 - bl L.4.23 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (7*2+1), %o2 - -L.4.23: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (7*2-1), %o2 - - -L.3.19: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 5 - bl L.4.21 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (5*2+1), %o2 - -L.4.21: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (5*2-1), %o2 - - - -L.2.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits 1 - bl L.3.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits 3 - bl L.4.19 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (3*2+1), %o2 - -L.4.19: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (3*2-1), %o2 - - -L.3.17: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits 1 - bl L.4.17 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (1*2+1), %o2 - -L.4.17: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (1*2-1), %o2 - - - - -L.1.16: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 2, accumulated bits -1 - bl L.2.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 3, accumulated bits -1 - bl L.3.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -1 - bl L.4.15 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2+1), %o2 - -L.4.15: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-1*2-1), %o2 - - -L.3.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -3 - bl L.4.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2+1), %o2 - -L.4.13: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-3*2-1), %o2 - - - -L.2.15: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 3, accumulated bits -3 - bl L.3.13 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - ! depth 4, accumulated bits -5 - bl L.4.11 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2+1), %o2 - -L.4.11: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-5*2-1), %o2 - - -L.3.13: - ! remainder is negative - addcc %o3,%o5,%o3 - ! depth 4, accumulated bits -7 - bl L.4.9 - srl %o5,1,%o5 - ! remainder is positive - subcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2+1), %o2 - -L.4.9: - ! remainder is negative - addcc %o3,%o5,%o3 - b 9f - add %o2, (-7*2-1), %o2 - - - - - 9: -Lend_regular_divide: - subcc %o4, 1, %o4 - bge Ldivloop - tst %o3 - bl,a Lgot_result - ! non-restoring fixup here (one instruction only!) - add %o3, %o1, %o3 - - -Lgot_result: - - retl - mov %o3, %o0 -- cgit 1.4.1